1792 lines
72 KiB
Python
1792 lines
72 KiB
Python
# ai_client.py
|
|
"""
|
|
Note(Gemini):
|
|
Acts as the unified interface for multiple LLM providers (Anthropic, Gemini).
|
|
Abstracts away the differences in how they handle tool schemas, history, and caching.
|
|
|
|
For Anthropic: aggressively manages the ~200k token limit by manually culling
|
|
stale [FILES UPDATED] entries and dropping the oldest message pairs.
|
|
|
|
For Gemini: injects the initial context directly into system_instruction
|
|
during chat creation to avoid massive history bloat.
|
|
"""
|
|
# ai_client.py
|
|
import tomllib
|
|
import json
|
|
import sys
|
|
import time
|
|
import datetime
|
|
import hashlib
|
|
import difflib
|
|
import threading
|
|
import requests
|
|
from pathlib import Path
|
|
from typing import Optional, Callable, Any
|
|
import os
|
|
import project_manager
|
|
import file_cache
|
|
import mcp_client
|
|
import anthropic
|
|
from gemini_cli_adapter import GeminiCliAdapter
|
|
from google import genai
|
|
from google.genai import types
|
|
from events import EventEmitter
|
|
|
|
_provider: str = "gemini"
|
|
_model: str = "gemini-2.5-flash-lite"
|
|
_temperature: float = 0.0
|
|
_max_tokens: int = 8192
|
|
|
|
_history_trunc_limit: int = 8000
|
|
|
|
# Global event emitter for API lifecycle events
|
|
events = EventEmitter()
|
|
|
|
def set_model_params(temp: float, max_tok: int, trunc_limit: int = 8000) -> None:
|
|
global _temperature, _max_tokens, _history_trunc_limit
|
|
_temperature = temp
|
|
_max_tokens = max_tok
|
|
_history_trunc_limit = trunc_limit
|
|
|
|
def get_history_trunc_limit() -> int:
|
|
return _history_trunc_limit
|
|
|
|
def set_history_trunc_limit(val: int) -> None:
|
|
global _history_trunc_limit
|
|
_history_trunc_limit = val
|
|
|
|
_gemini_client = None
|
|
_gemini_chat = None
|
|
_gemini_cache = None
|
|
_gemini_cache_md_hash: int | None = None
|
|
_gemini_cache_created_at: float | None = None
|
|
|
|
# Gemini cache TTL in seconds. Caches are created with this TTL and
|
|
# proactively rebuilt at 90% of this value to avoid stale-reference errors.
|
|
_GEMINI_CACHE_TTL = 3600
|
|
|
|
_anthropic_client = None
|
|
_anthropic_history: list[dict] = []
|
|
_anthropic_history_lock = threading.Lock()
|
|
|
|
_deepseek_client = None
|
|
_deepseek_history: list[dict] = []
|
|
_deepseek_history_lock = threading.Lock()
|
|
|
|
_send_lock = threading.Lock()
|
|
|
|
_gemini_cli_adapter = None
|
|
|
|
# Injected by gui.py - called when AI wants to run a command.
|
|
# Signature: (script: str, base_dir: str) -> str | None
|
|
confirm_and_run_callback = None
|
|
|
|
# Injected by gui.py - called whenever a comms entry is appended.
|
|
# Signature: (entry: dict) -> None
|
|
comms_log_callback = None
|
|
|
|
# Injected by gui.py - called whenever a tool call completes.
|
|
# Signature: (script: str, result: str) -> None
|
|
tool_log_callback = None
|
|
|
|
# Increased to allow thorough code exploration before forcing a summary
|
|
MAX_TOOL_ROUNDS = 10
|
|
|
|
# Maximum cumulative bytes of tool output allowed per send() call.
|
|
# Prevents unbounded memory growth during long tool-calling loops.
|
|
_MAX_TOOL_OUTPUT_BYTES = 500_000
|
|
|
|
# Maximum characters per text chunk sent to Anthropic.
|
|
# Kept well under the ~200k token API limit.
|
|
_ANTHROPIC_CHUNK_SIZE = 120_000
|
|
|
|
_SYSTEM_PROMPT = (
|
|
"You are a helpful coding assistant with access to a PowerShell tool and MCP tools (file access: read_file, list_directory, search_files, get_file_summary, web access: web_search, fetch_url). "
|
|
"When calling file/directory tools, always use the 'path' parameter for the target path. "
|
|
"When asked to create or edit files, prefer targeted edits over full rewrites. "
|
|
"Always explain what you are doing before invoking the tool.\n\n"
|
|
"When writing or rewriting large files (especially those containing quotes, backticks, or special characters), "
|
|
"avoid python -c with inline strings. Instead: (1) write a .py helper script to disk using a PS here-string "
|
|
"(@'...'@ for literal content), (2) run it with `python <script>`, (3) delete the helper. "
|
|
"For small targeted edits, use PowerShell's (Get-Content) / .Replace() / Set-Content or Add-Content directly.\n\n"
|
|
"When making function calls using tools that accept array or object parameters "
|
|
"ensure those are structured using JSON. For example:\n"
|
|
"When you need to verify a change, rely on the exit code and stdout/stderr from the tool \u2014 "
|
|
"the user's context files are automatically refreshed after every tool call, so you do NOT "
|
|
"need to re-read files that are already provided in the <context> block."
|
|
)
|
|
|
|
_custom_system_prompt: str = ""
|
|
|
|
def set_custom_system_prompt(prompt: str) -> None:
|
|
global _custom_system_prompt
|
|
_custom_system_prompt = prompt
|
|
|
|
def _get_combined_system_prompt() -> str:
|
|
if _custom_system_prompt.strip():
|
|
return f"{_SYSTEM_PROMPT}\n\n[USER SYSTEM PROMPT]\n{_custom_system_prompt}"
|
|
return _SYSTEM_PROMPT
|
|
# ------------------------------------------------------------------ comms log
|
|
|
|
_comms_log: list[dict] = []
|
|
|
|
COMMS_CLAMP_CHARS = 300
|
|
|
|
def _append_comms(direction: str, kind: str, payload: dict[str, Any]) -> None:
|
|
entry = {
|
|
"ts": datetime.datetime.now().strftime("%H:%M:%S"),
|
|
"direction": direction,
|
|
"kind": kind,
|
|
"provider": _provider,
|
|
"model": _model,
|
|
"payload": payload,
|
|
}
|
|
_comms_log.append(entry)
|
|
if comms_log_callback is not None:
|
|
comms_log_callback(entry)
|
|
|
|
def get_comms_log() -> list[dict]:
|
|
return list(_comms_log)
|
|
|
|
def clear_comms_log() -> None:
|
|
_comms_log.clear()
|
|
|
|
def _load_credentials() -> dict:
|
|
cred_path = os.environ.get("SLOP_CREDENTIALS", "credentials.toml")
|
|
try:
|
|
with open(cred_path, "rb") as f:
|
|
return tomllib.load(f)
|
|
except FileNotFoundError:
|
|
raise FileNotFoundError(
|
|
f"Credentials file not found: {cred_path}\n"
|
|
f"Create a credentials.toml with:\n"
|
|
f" [gemini]\n api_key = \"your-key\"\n"
|
|
f" [anthropic]\n api_key = \"your-key\"\n"
|
|
f" [deepseek]\n api_key = \"your-key\"\n"
|
|
f"Or set SLOP_CREDENTIALS env var to a custom path."
|
|
)
|
|
# ------------------------------------------------------------------ provider errors
|
|
|
|
class ProviderError(Exception):
|
|
def __init__(self, kind: str, provider: str, original: Exception) -> None:
|
|
self.kind = kind
|
|
self.provider = provider
|
|
self.original = original
|
|
super().__init__(str(original))
|
|
|
|
def ui_message(self) -> str:
|
|
labels = {
|
|
"quota": "QUOTA EXHAUSTED",
|
|
"rate_limit": "RATE LIMITED",
|
|
"auth": "AUTH / API KEY ERROR",
|
|
"balance": "BALANCE / BILLING ERROR",
|
|
"network": "NETWORK / CONNECTION ERROR",
|
|
"unknown": "API ERROR",
|
|
}
|
|
label = labels.get(self.kind, "API ERROR")
|
|
return f"[{self.provider.upper()} {label}]\n\n{self.original}"
|
|
|
|
def _classify_anthropic_error(exc: Exception) -> ProviderError:
|
|
try:
|
|
if isinstance(exc, anthropic.RateLimitError):
|
|
return ProviderError("rate_limit", "anthropic", exc)
|
|
if isinstance(exc, anthropic.AuthenticationError):
|
|
return ProviderError("auth", "anthropic", exc)
|
|
if isinstance(exc, anthropic.PermissionDeniedError):
|
|
return ProviderError("auth", "anthropic", exc)
|
|
if isinstance(exc, anthropic.APIConnectionError):
|
|
return ProviderError("network", "anthropic", exc)
|
|
if isinstance(exc, anthropic.APIStatusError):
|
|
status = getattr(exc, "status_code", 0)
|
|
body = str(exc).lower()
|
|
if status == 429:
|
|
return ProviderError("rate_limit", "anthropic", exc)
|
|
if status in (401, 403):
|
|
return ProviderError("auth", "anthropic", exc)
|
|
if status == 402:
|
|
return ProviderError("balance", "anthropic", exc)
|
|
if "credit" in body or "balance" in body or "billing" in body:
|
|
return ProviderError("balance", "anthropic", exc)
|
|
if "quota" in body or "limit" in body or "exceeded" in body:
|
|
return ProviderError("quota", "anthropic", exc)
|
|
except ImportError:
|
|
pass
|
|
return ProviderError("unknown", "anthropic", exc)
|
|
|
|
def _classify_gemini_error(exc: Exception) -> ProviderError:
|
|
body = str(exc).lower()
|
|
try:
|
|
from google.api_core import exceptions as gac
|
|
if isinstance(exc, gac.ResourceExhausted):
|
|
return ProviderError("quota", "gemini", exc)
|
|
if isinstance(exc, gac.TooManyRequests):
|
|
return ProviderError("rate_limit", "gemini", exc)
|
|
if isinstance(exc, (gac.Unauthenticated, gac.PermissionDenied)):
|
|
return ProviderError("auth", "gemini", exc)
|
|
if isinstance(exc, gac.ServiceUnavailable):
|
|
return ProviderError("network", "gemini", exc)
|
|
except ImportError:
|
|
pass
|
|
if "429" in body or "quota" in body or "resource exhausted" in body:
|
|
return ProviderError("quota", "gemini", exc)
|
|
if "rate" in body and "limit" in body:
|
|
return ProviderError("rate_limit", "gemini", exc)
|
|
if "401" in body or "403" in body or "api key" in body or "unauthenticated" in body:
|
|
return ProviderError("auth", "gemini", exc)
|
|
if "402" in body or "billing" in body or "balance" in body or "payment" in body:
|
|
return ProviderError("balance", "gemini", exc)
|
|
if "connection" in body or "timeout" in body or "unreachable" in body:
|
|
return ProviderError("network", "gemini", exc)
|
|
return ProviderError("unknown", "gemini", exc)
|
|
|
|
def _classify_deepseek_error(exc: Exception) -> ProviderError:
|
|
body = str(exc).lower()
|
|
if "429" in body or "rate" in body:
|
|
return ProviderError("rate_limit", "deepseek", exc)
|
|
if "401" in body or "403" in body or "auth" in body or "api key" in body:
|
|
return ProviderError("auth", "deepseek", exc)
|
|
if "402" in body or "balance" in body or "billing" in body:
|
|
return ProviderError("balance", "deepseek", exc)
|
|
if "quota" in body or "limit exceeded" in body:
|
|
return ProviderError("quota", "deepseek", exc)
|
|
if "connection" in body or "timeout" in body or "network" in body:
|
|
return ProviderError("network", "deepseek", exc)
|
|
return ProviderError("unknown", "deepseek", exc)
|
|
# ------------------------------------------------------------------ provider setup
|
|
|
|
def set_provider(provider: str, model: str) -> None:
|
|
global _provider, _model
|
|
_provider = provider
|
|
if provider == "gemini_cli":
|
|
valid_models = _list_gemini_cli_models()
|
|
# If model is invalid or belongs to another provider (like deepseek), force default
|
|
if model not in valid_models or model.startswith("deepseek"):
|
|
_model = "gemini-3-flash-preview"
|
|
else:
|
|
_model = model
|
|
else:
|
|
_model = model
|
|
|
|
def cleanup() -> None:
|
|
"""Called on application exit to prevent orphaned caches from billing."""
|
|
global _gemini_client, _gemini_cache
|
|
if _gemini_client and _gemini_cache:
|
|
try:
|
|
_gemini_client.caches.delete(name=_gemini_cache.name)
|
|
except Exception:
|
|
pass
|
|
|
|
def reset_session() -> None:
|
|
global _gemini_client, _gemini_chat, _gemini_cache
|
|
global _gemini_cache_md_hash, _gemini_cache_created_at
|
|
global _anthropic_client, _anthropic_history
|
|
global _CACHED_ANTHROPIC_TOOLS
|
|
global _gemini_cli_adapter
|
|
if _gemini_client and _gemini_cache:
|
|
try:
|
|
_gemini_client.caches.delete(name=_gemini_cache.name)
|
|
except Exception:
|
|
pass
|
|
_gemini_client = None
|
|
_gemini_chat = None
|
|
_gemini_cache = None
|
|
_gemini_cache_md_hash = None
|
|
_gemini_cache_created_at = None
|
|
if _gemini_cli_adapter:
|
|
_gemini_cli_adapter.session_id = None
|
|
_gemini_cli_adapter = None
|
|
_anthropic_client = None
|
|
with _anthropic_history_lock:
|
|
_anthropic_history = []
|
|
_deepseek_client = None
|
|
with _deepseek_history_lock:
|
|
_deepseek_history = []
|
|
_CACHED_ANTHROPIC_TOOLS = None
|
|
file_cache.reset_client()
|
|
|
|
def get_gemini_cache_stats() -> dict[str, Any]:
|
|
"""
|
|
Retrieves statistics about the Gemini caches, such as count and total size.
|
|
"""
|
|
_ensure_gemini_client()
|
|
caches_iterator = _gemini_client.caches.list()
|
|
caches = list(caches_iterator)
|
|
total_size_bytes = sum(c.size_bytes for c in caches)
|
|
return {
|
|
"cache_count": len(list(caches)),
|
|
"total_size_bytes": total_size_bytes,
|
|
}
|
|
# ------------------------------------------------------------------ model listing
|
|
|
|
def list_models(provider: str) -> list[str]:
|
|
creds = _load_credentials()
|
|
if provider == "gemini":
|
|
return _list_gemini_models(creds["gemini"]["api_key"])
|
|
elif provider == "anthropic":
|
|
return _list_anthropic_models()
|
|
elif provider == "deepseek":
|
|
return _list_deepseek_models(creds["deepseek"]["api_key"])
|
|
elif provider == "gemini_cli":
|
|
return _list_gemini_cli_models()
|
|
return []
|
|
|
|
def _list_gemini_cli_models() -> list[str]:
|
|
"""
|
|
List available Gemini models for the CLI.
|
|
Since the CLI doesn't have a direct 'list models' command yet,
|
|
we return a curated list of supported models based on CLI metadata.
|
|
"""
|
|
return [
|
|
"gemini-3-flash-preview",
|
|
"gemini-3.1-pro-preview",
|
|
"gemini-2.5-pro",
|
|
"gemini-2.5-flash",
|
|
"gemini-2.5-flash-lite",
|
|
]
|
|
|
|
def _list_gemini_models(api_key: str) -> list[str]:
|
|
|
|
try:
|
|
client = genai.Client(api_key=api_key)
|
|
models = []
|
|
for m in client.models.list():
|
|
name = m.name
|
|
if name.startswith("models/"):
|
|
name = name[len("models/"):]
|
|
if "gemini" in name.lower():
|
|
models.append(name)
|
|
return sorted(models)
|
|
except Exception as exc:
|
|
raise _classify_gemini_error(exc) from exc
|
|
|
|
def _list_anthropic_models() -> list[str]:
|
|
|
|
try:
|
|
creds = _load_credentials()
|
|
client = anthropic.Anthropic(api_key=creds["anthropic"]["api_key"])
|
|
models = []
|
|
for m in client.models.list():
|
|
models.append(m.id)
|
|
return sorted(models)
|
|
except Exception as exc:
|
|
raise _classify_anthropic_error(exc) from exc
|
|
|
|
def _list_deepseek_models(api_key: str) -> list[str]:
|
|
"""
|
|
List available DeepSeek models.
|
|
"""
|
|
# For now, return the models specified in the requirements
|
|
return ["deepseek-chat", "deepseek-reasoner", "deepseek-v3", "deepseek-r1"]
|
|
# ------------------------------------------------------------------ tool definition
|
|
|
|
TOOL_NAME = "run_powershell"
|
|
|
|
_agent_tools: dict = {}
|
|
|
|
def set_agent_tools(tools: dict[str, bool]) -> None:
|
|
global _agent_tools, _CACHED_ANTHROPIC_TOOLS
|
|
_agent_tools = tools
|
|
_CACHED_ANTHROPIC_TOOLS = None
|
|
|
|
def _build_anthropic_tools() -> list[dict]:
|
|
"""Build the full Anthropic tools list: run_powershell + MCP file tools."""
|
|
mcp_tools = []
|
|
for spec in mcp_client.MCP_TOOL_SPECS:
|
|
if _agent_tools.get(spec["name"], True):
|
|
mcp_tools.append({
|
|
"name": spec["name"],
|
|
"description": spec["description"],
|
|
"input_schema": spec["parameters"],
|
|
})
|
|
tools_list = mcp_tools
|
|
if _agent_tools.get(TOOL_NAME, True):
|
|
powershell_tool = {
|
|
"name": TOOL_NAME,
|
|
"description": (
|
|
"Run a PowerShell script within the project base_dir. "
|
|
"Use this to create, edit, rename, or delete files and directories. "
|
|
"The working directory is set to base_dir automatically. "
|
|
"Always prefer targeted edits over full rewrites where possible. "
|
|
"stdout and stderr are returned to you as the result."
|
|
),
|
|
"input_schema": {
|
|
"type": "object",
|
|
"properties": {
|
|
"script": {
|
|
"type": "string",
|
|
"description": "The PowerShell script to execute."
|
|
}
|
|
},
|
|
"required": ["script"]
|
|
},
|
|
"cache_control": {"type": "ephemeral"},
|
|
}
|
|
tools_list.append(powershell_tool)
|
|
elif tools_list:
|
|
# Anthropic requires the LAST tool to have cache_control for the prefix caching to work optimally on tools
|
|
tools_list[-1]["cache_control"] = {"type": "ephemeral"}
|
|
return tools_list
|
|
|
|
_ANTHROPIC_TOOLS = _build_anthropic_tools()
|
|
|
|
_CACHED_ANTHROPIC_TOOLS = None
|
|
|
|
def _get_anthropic_tools() -> list[dict]:
|
|
"""Return the Anthropic tools list, rebuilding only once per session."""
|
|
global _CACHED_ANTHROPIC_TOOLS
|
|
if _CACHED_ANTHROPIC_TOOLS is None:
|
|
_CACHED_ANTHROPIC_TOOLS = _build_anthropic_tools()
|
|
return _CACHED_ANTHROPIC_TOOLS
|
|
|
|
def _gemini_tool_declaration() -> types.Tool | None:
|
|
declarations = []
|
|
# MCP file tools
|
|
for spec in mcp_client.MCP_TOOL_SPECS:
|
|
if not _agent_tools.get(spec["name"], True):
|
|
continue
|
|
props = {}
|
|
for pname, pdef in spec["parameters"].get("properties", {}).items():
|
|
ptype_str = pdef.get("type", "string").upper()
|
|
ptype = getattr(types.Type, ptype_str, types.Type.STRING)
|
|
props[pname] = types.Schema(
|
|
type=ptype,
|
|
description=pdef.get("description", ""),
|
|
)
|
|
declarations.append(types.FunctionDeclaration(
|
|
name=spec["name"],
|
|
description=spec["description"],
|
|
parameters=types.Schema(
|
|
type=types.Type.OBJECT,
|
|
properties=props,
|
|
required=spec["parameters"].get("required", []),
|
|
),
|
|
))
|
|
# PowerShell tool
|
|
if _agent_tools.get(TOOL_NAME, True):
|
|
declarations.append(types.FunctionDeclaration(
|
|
name=TOOL_NAME,
|
|
description=(
|
|
"Run a PowerShell script within the project base_dir. "
|
|
"Use this to create, edit, rename, or delete files and directories. "
|
|
"The working directory is set to base_dir automatically. "
|
|
"stdout and stderr are returned to you as the result."
|
|
),
|
|
parameters=types.Schema(
|
|
type=types.Type.OBJECT,
|
|
properties={
|
|
"script": types.Schema(
|
|
type=types.Type.STRING,
|
|
description="The PowerShell script to execute."
|
|
)
|
|
},
|
|
required=["script"]
|
|
),
|
|
))
|
|
return types.Tool(function_declarations=declarations) if declarations else None
|
|
|
|
def _run_script(script: str, base_dir: str, qa_callback: Optional[Callable[[str], str]] = None) -> str:
|
|
if confirm_and_run_callback is None:
|
|
return "ERROR: no confirmation handler registered"
|
|
result = confirm_and_run_callback(script, base_dir, qa_callback)
|
|
if result is None:
|
|
output = "USER REJECTED: command was not executed"
|
|
else:
|
|
output = result
|
|
if tool_log_callback is not None:
|
|
tool_log_callback(script, output)
|
|
return output
|
|
|
|
def _truncate_tool_output(output: str) -> str:
|
|
"""Truncate tool output to _history_trunc_limit chars before sending to API."""
|
|
if _history_trunc_limit > 0 and len(output) > _history_trunc_limit:
|
|
return output[:_history_trunc_limit] + "\n\n... [TRUNCATED BY SYSTEM TO SAVE TOKENS.]"
|
|
return output
|
|
# ------------------------------------------------------------------ dynamic file context refresh
|
|
|
|
def _reread_file_items(file_items: list[dict]) -> tuple[list[dict], list[dict]]:
|
|
"""
|
|
Re-read file_items from disk, but only files whose mtime has changed.
|
|
Returns (all_items, changed_items) — all_items is the full refreshed list,
|
|
changed_items contains only the files that were actually modified since
|
|
the last read (used to build a minimal [FILES UPDATED] block).
|
|
"""
|
|
refreshed = []
|
|
changed = []
|
|
for item in file_items:
|
|
path = item.get("path")
|
|
if path is None:
|
|
refreshed.append(item)
|
|
continue
|
|
from pathlib import Path as _P
|
|
p = _P(path) if not isinstance(path, _P) else path
|
|
try:
|
|
current_mtime = p.stat().st_mtime
|
|
prev_mtime = item.get("mtime", 0.0)
|
|
if current_mtime == prev_mtime:
|
|
refreshed.append(item) # unchanged — skip re-read
|
|
continue
|
|
content = p.read_text(encoding="utf-8")
|
|
new_item = {**item, "old_content": item.get("content", ""), "content": content, "error": False, "mtime": current_mtime}
|
|
refreshed.append(new_item)
|
|
changed.append(new_item)
|
|
except Exception as e:
|
|
err_item = {**item, "content": f"ERROR re-reading {p}: {e}", "error": True, "mtime": 0.0}
|
|
refreshed.append(err_item)
|
|
changed.append(err_item)
|
|
return refreshed, changed
|
|
|
|
def _build_file_context_text(file_items: list[dict]) -> str:
|
|
"""
|
|
Build a compact text summary of all files from file_items, suitable for
|
|
injecting into a tool_result message so the AI sees current file contents.
|
|
"""
|
|
if not file_items:
|
|
return ""
|
|
parts = []
|
|
for item in file_items:
|
|
path = item.get("path") or item.get("entry", "unknown")
|
|
suffix = str(path).rsplit(".", 1)[-1] if "." in str(path) else "text"
|
|
content = item.get("content", "")
|
|
parts.append(f"### `{path}`\n\n```{suffix}\n{content}\n```")
|
|
return "\n\n---\n\n".join(parts)
|
|
|
|
_DIFF_LINE_THRESHOLD = 200
|
|
|
|
def _build_file_diff_text(changed_items: list[dict]) -> str:
|
|
"""
|
|
Build text for changed files. Small files (<= _DIFF_LINE_THRESHOLD lines)
|
|
get full content; large files get a unified diff against old_content.
|
|
"""
|
|
if not changed_items:
|
|
return ""
|
|
parts = []
|
|
for item in changed_items:
|
|
path = item.get("path") or item.get("entry", "unknown")
|
|
content = item.get("content", "")
|
|
old_content = item.get("old_content", "")
|
|
new_lines = content.splitlines(keepends=True)
|
|
if len(new_lines) <= _DIFF_LINE_THRESHOLD or not old_content:
|
|
suffix = str(path).rsplit(".", 1)[-1] if "." in str(path) else "text"
|
|
parts.append(f"### `{path}` (full)\n\n```{suffix}\n{content}\n```")
|
|
else:
|
|
old_lines = old_content.splitlines(keepends=True)
|
|
diff = difflib.unified_diff(old_lines, new_lines, fromfile=str(path), tofile=str(path), lineterm="")
|
|
diff_text = "\n".join(diff)
|
|
if diff_text:
|
|
parts.append(f"### `{path}` (diff)\n\n```diff\n{diff_text}\n```")
|
|
else:
|
|
parts.append(f"### `{path}` (no changes detected)")
|
|
return "\n\n---\n\n".join(parts)
|
|
# ------------------------------------------------------------------ content block serialisation
|
|
|
|
def _content_block_to_dict(block: Any) -> dict[str, Any]:
|
|
"""
|
|
Convert an Anthropic SDK content block object to a plain dict.
|
|
This ensures history entries are always JSON-serialisable dicts,
|
|
not opaque SDK objects that may fail on re-serialisation.
|
|
"""
|
|
if isinstance(block, dict):
|
|
return block
|
|
if hasattr(block, "model_dump"):
|
|
return block.model_dump()
|
|
if hasattr(block, "to_dict"):
|
|
return block.to_dict()
|
|
# Fallback: manually construct based on type
|
|
block_type = getattr(block, "type", None)
|
|
if block_type == "text":
|
|
return {"type": "text", "text": block.text}
|
|
if block_type == "tool_use":
|
|
return {"type": "tool_use", "id": block.id, "name": block.name, "input": block.input}
|
|
return {"type": "text", "text": str(block)}
|
|
# ------------------------------------------------------------------ gemini
|
|
|
|
def _ensure_gemini_client() -> None:
|
|
global _gemini_client
|
|
if _gemini_client is None:
|
|
creds = _load_credentials()
|
|
_gemini_client = genai.Client(api_key=creds["gemini"]["api_key"])
|
|
|
|
def _get_gemini_history_list(chat: Any | None) -> list[Any]:
|
|
if not chat: return []
|
|
# google-genai SDK stores the mutable list in _history
|
|
if hasattr(chat, "_history"):
|
|
return chat._history
|
|
if hasattr(chat, "history"):
|
|
return chat.history
|
|
if hasattr(chat, "get_history"):
|
|
return chat.get_history()
|
|
return []
|
|
|
|
def _send_gemini(md_content: str, user_message: str, base_dir: str,
|
|
file_items: list[dict[str, Any]] | None = None,
|
|
discussion_history: str = "",
|
|
pre_tool_callback: Optional[Callable[[str], bool]] = None,
|
|
qa_callback: Optional[Callable[[str], str]] = None) -> str:
|
|
global _gemini_chat, _gemini_cache, _gemini_cache_md_hash, _gemini_cache_created_at
|
|
try:
|
|
_ensure_gemini_client(); mcp_client.configure(file_items or [], [base_dir])
|
|
# Only stable content (files + screenshots) goes in the cached system instruction.
|
|
# Discussion history is sent as conversation messages so the cache isn't invalidated every turn.
|
|
sys_instr = f"{_get_combined_system_prompt()}\n\n<context>\n{md_content}\n</context>"
|
|
td = _gemini_tool_declaration()
|
|
tools_decl = [td] if td else None
|
|
# DYNAMIC CONTEXT: Check if files/context changed mid-session
|
|
current_md_hash = hashlib.md5(md_content.encode()).hexdigest()
|
|
old_history = None
|
|
if _gemini_chat and _gemini_cache_md_hash != current_md_hash:
|
|
old_history = list(_get_gemini_history_list(_gemini_chat)) if _get_gemini_history_list(_gemini_chat) else []
|
|
if _gemini_cache:
|
|
try: _gemini_client.caches.delete(name=_gemini_cache.name)
|
|
except Exception as e: _append_comms("OUT", "request", {"message": f"[CACHE DELETE WARN] {e}"})
|
|
_gemini_chat = None
|
|
_gemini_cache = None
|
|
_gemini_cache_created_at = None
|
|
_append_comms("OUT", "request", {"message": "[CONTEXT CHANGED] Rebuilding cache and chat session..."})
|
|
# CACHE TTL: Proactively rebuild before the cache expires server-side.
|
|
# If we don't, send_message() will reference a deleted cache and fail.
|
|
if _gemini_chat and _gemini_cache and _gemini_cache_created_at:
|
|
elapsed = time.time() - _gemini_cache_created_at
|
|
if elapsed > _GEMINI_CACHE_TTL * 0.9:
|
|
old_history = list(_get_gemini_history_list(_gemini_chat)) if _get_gemini_history_list(_get_gemini_history_list(_gemini_chat)) else []
|
|
try: _gemini_client.caches.delete(name=_gemini_cache.name)
|
|
except Exception as e: _append_comms("OUT", "request", {"message": f"[CACHE DELETE WARN] {e}"})
|
|
_gemini_chat = None
|
|
_gemini_cache = None
|
|
_gemini_cache_created_at = None
|
|
_append_comms("OUT", "request", {"message": f"[CACHE TTL] Rebuilding cache (expired after {int(elapsed)}s)..."})
|
|
if not _gemini_chat:
|
|
chat_config = types.GenerateContentConfig(
|
|
system_instruction=sys_instr,
|
|
tools=tools_decl,
|
|
temperature=_temperature,
|
|
max_output_tokens=_max_tokens,
|
|
safety_settings=[types.SafetySetting(category="HARM_CATEGORY_DANGEROUS_CONTENT", threshold="BLOCK_ONLY_HIGH")]
|
|
)
|
|
# Check if context is large enough to warrant caching (min 2048 tokens usually)
|
|
should_cache = False
|
|
try:
|
|
count_resp = _gemini_client.models.count_tokens(model=_model, contents=[sys_instr])
|
|
# We use a 2048 threshold to be safe across models
|
|
if count_resp.total_tokens >= 2048:
|
|
should_cache = True
|
|
else:
|
|
_append_comms("OUT", "request", {"message": f"[CACHING SKIPPED] Context too small ({count_resp.total_tokens} tokens < 2048)"})
|
|
except Exception as e:
|
|
_append_comms("OUT", "request", {"message": f"[COUNT FAILED] {e}"})
|
|
if should_cache:
|
|
try:
|
|
# Gemini requires 1024 (Flash) or 4096 (Pro) tokens to cache.
|
|
_gemini_cache = _gemini_client.caches.create(
|
|
model=_model,
|
|
config=types.CreateCachedContentConfig(
|
|
system_instruction=sys_instr,
|
|
tools=tools_decl,
|
|
ttl=f"{_GEMINI_CACHE_TTL}s",
|
|
)
|
|
)
|
|
_gemini_cache_created_at = time.time()
|
|
chat_config = types.GenerateContentConfig(
|
|
cached_content=_gemini_cache.name,
|
|
temperature=_temperature,
|
|
max_output_tokens=_max_tokens,
|
|
safety_settings=[types.SafetySetting(category="HARM_CATEGORY_DANGEROUS_CONTENT", threshold="BLOCK_ONLY_HIGH")]
|
|
)
|
|
_append_comms("OUT", "request", {"message": f"[CACHE CREATED] {_gemini_cache.name}"})
|
|
except Exception as e:
|
|
_gemini_cache = None
|
|
_gemini_cache_created_at = None
|
|
_append_comms("OUT", "request", {"message": f"[CACHE FAILED] {type(e).__name__}: {e} \u2014 falling back to inline system_instruction"})
|
|
kwargs = {"model": _model, "config": chat_config}
|
|
if old_history:
|
|
kwargs["history"] = old_history
|
|
_gemini_chat = _gemini_client.chats.create(**kwargs)
|
|
_gemini_cache_md_hash = current_md_hash
|
|
# Inject discussion history as a user message on first chat creation
|
|
# (only when there's no old_history being restored, i.e., fresh session)
|
|
if discussion_history and not old_history:
|
|
_gemini_chat.send_message(f"[DISCUSSION HISTORY]\n\n{discussion_history}")
|
|
_append_comms("OUT", "request", {"message": f"[HISTORY INJECTED] {len(discussion_history)} chars"})
|
|
_append_comms("OUT", "request", {"message": f"[ctx {len(md_content)} + msg {len(user_message)}]"})
|
|
payload: str | list[types.Part] = user_message
|
|
all_text: list[str] = []
|
|
_cumulative_tool_bytes = 0
|
|
# Strip stale file refreshes and truncate old tool outputs ONCE before
|
|
# entering the tool loop (not per-round \u2014 history entries don't change).
|
|
if _gemini_chat and _get_gemini_history_list(_gemini_chat):
|
|
for msg in _get_gemini_history_list(_gemini_chat):
|
|
if msg.role == "user" and hasattr(msg, "parts"):
|
|
for p in msg.parts:
|
|
if hasattr(p, "function_response") and p.function_response and hasattr(p.function_response, "response"):
|
|
r = p.function_response.response
|
|
if isinstance(r, dict) and "output" in r:
|
|
val = r["output"]
|
|
if isinstance(val, str):
|
|
if "[SYSTEM: FILES UPDATED]" in val:
|
|
val = val.split("[SYSTEM: FILES UPDATED]")[0].strip()
|
|
if _history_trunc_limit > 0 and len(val) > _history_trunc_limit:
|
|
val = val[:_history_trunc_limit] + "\n\n... [TRUNCATED BY SYSTEM TO SAVE TOKENS.]"
|
|
r["output"] = val
|
|
for r_idx in range(MAX_TOOL_ROUNDS + 2):
|
|
events.emit("request_start", payload={"provider": "gemini", "model": _model, "round": r_idx})
|
|
resp = _gemini_chat.send_message(payload)
|
|
txt = "\n".join(p.text for c in resp.candidates if getattr(c, "content", None) for p in c.content.parts if hasattr(p, "text") and p.text)
|
|
if txt: all_text.append(txt)
|
|
calls = [p.function_call for c in resp.candidates if getattr(c, "content", None) for p in c.content.parts if hasattr(p, "function_call") and p.function_call]
|
|
usage = {"input_tokens": getattr(resp.usage_metadata, "prompt_token_count", 0), "output_tokens": getattr(resp.usage_metadata, "candidates_token_count", 0)}
|
|
cached_tokens = getattr(resp.usage_metadata, "cached_content_token_count", None)
|
|
if cached_tokens:
|
|
usage["cache_read_input_tokens"] = cached_tokens
|
|
events.emit("response_received", payload={"provider": "gemini", "model": _model, "usage": usage, "round": r_idx})
|
|
reason = resp.candidates[0].finish_reason.name if resp.candidates and hasattr(resp.candidates[0], "finish_reason") else "STOP"
|
|
_append_comms("IN", "response", {"round": r_idx, "stop_reason": reason, "text": txt, "tool_calls": [{"name": c.name, "args": dict(c.args)} for c in calls], "usage": usage})
|
|
# Guard: proactively trim history when input tokens exceed 40% of limit
|
|
total_in = usage.get("input_tokens", 0)
|
|
if total_in > _GEMINI_MAX_INPUT_TOKENS * 0.4 and _gemini_chat and _get_gemini_history_list(_gemini_chat):
|
|
hist = _get_gemini_history_list(_gemini_chat)
|
|
dropped = 0
|
|
# Drop oldest pairs (user+model) but keep at least the last 2 entries
|
|
while len(hist) > 4 and total_in > _GEMINI_MAX_INPUT_TOKENS * 0.3:
|
|
# Drop in pairs (user + model) to maintain alternating roles required by Gemini
|
|
saved = 0
|
|
for _ in range(2):
|
|
if not hist: break
|
|
for p in hist[0].parts:
|
|
if hasattr(p, "text") and p.text:
|
|
saved += int(len(p.text) / _CHARS_PER_TOKEN)
|
|
elif hasattr(p, "function_response") and p.function_response:
|
|
r = getattr(p.function_response, "response", {})
|
|
if isinstance(r, dict):
|
|
saved += int(len(str(r.get("output", ""))) / _CHARS_PER_TOKEN)
|
|
hist.pop(0)
|
|
dropped += 1
|
|
total_in -= max(saved, 200)
|
|
if dropped > 0:
|
|
_append_comms("OUT", "request", {"message": f"[GEMINI HISTORY TRIMMED: dropped {dropped} old entries to stay within token budget]"})
|
|
if not calls or r_idx > MAX_TOOL_ROUNDS: break
|
|
f_resps: list[types.Part] = []
|
|
log: list[dict[str, Any]] = []
|
|
for i, fc in enumerate(calls):
|
|
name, args = fc.name, dict(fc.args)
|
|
# Check for tool confirmation if callback is provided
|
|
if pre_tool_callback:
|
|
payload_str = json.dumps({"tool": name, "args": args})
|
|
if not pre_tool_callback(payload_str):
|
|
out = "USER REJECTED: tool execution cancelled"
|
|
f_resps.append(types.Part.from_function_response(name=name, response={"output": out}))
|
|
log.append({"tool_use_id": name, "content": out})
|
|
continue
|
|
events.emit("tool_execution", payload={"status": "started", "tool": name, "args": args, "round": r_idx})
|
|
if name in mcp_client.TOOL_NAMES:
|
|
_append_comms("OUT", "tool_call", {"name": name, "args": args})
|
|
out = mcp_client.dispatch(name, args)
|
|
elif name == TOOL_NAME:
|
|
scr = args.get("script", "")
|
|
_append_comms("OUT", "tool_call", {"name": TOOL_NAME, "script": scr})
|
|
out = _run_script(scr, base_dir, qa_callback)
|
|
else: out = f"ERROR: unknown tool '{name}'"
|
|
if i == len(calls) - 1:
|
|
if file_items:
|
|
file_items, changed = _reread_file_items(file_items)
|
|
ctx = _build_file_diff_text(changed)
|
|
if ctx:
|
|
out += f"\n\n[SYSTEM: FILES UPDATED]\n\n{ctx}"
|
|
if r_idx == MAX_TOOL_ROUNDS: out += "\n\n[SYSTEM: MAX ROUNDS. PROVIDE FINAL ANSWER.]"
|
|
out = _truncate_tool_output(out)
|
|
_cumulative_tool_bytes += len(out)
|
|
f_resps.append(types.Part.from_function_response(name=name, response={"output": out}))
|
|
log.append({"tool_use_id": name, "content": out})
|
|
events.emit("tool_execution", payload={"status": "completed", "tool": name, "result": out, "round": r_idx})
|
|
if _cumulative_tool_bytes > _MAX_TOOL_OUTPUT_BYTES:
|
|
f_resps.append(types.Part.from_text(
|
|
f"SYSTEM WARNING: Cumulative tool output exceeded {_MAX_TOOL_OUTPUT_BYTES // 1000}KB budget. Provide your final answer now."
|
|
))
|
|
_append_comms("OUT", "request", {"message": f"[TOOL OUTPUT BUDGET EXCEEDED: {_cumulative_tool_bytes} bytes]"})
|
|
_append_comms("OUT", "tool_result_send", {"results": log})
|
|
payload = f_resps
|
|
return "\n\n".join(all_text) if all_text else "(No text returned)"
|
|
except Exception as e: raise _classify_gemini_error(e) from e
|
|
|
|
def _send_gemini_cli(md_content: str, user_message: str, base_dir: str,
|
|
file_items: list[dict[str, Any]] | None = None,
|
|
discussion_history: str = "",
|
|
pre_tool_callback: Optional[Callable[[str], bool]] = None,
|
|
qa_callback: Optional[Callable[[str], str]] = None) -> str:
|
|
global _gemini_cli_adapter
|
|
try:
|
|
if _gemini_cli_adapter is None:
|
|
_gemini_cli_adapter = GeminiCliAdapter(binary_path="gemini")
|
|
adapter = _gemini_cli_adapter
|
|
mcp_client.configure(file_items or [], [base_dir])
|
|
# Construct the system instruction, combining the base system prompt and the current context.
|
|
sys_instr = f"{_get_combined_system_prompt()}\n\n<context>\n{md_content}\n</context>"
|
|
safety_settings = [{'category': 'HARM_CATEGORY_DANGEROUS_CONTENT', 'threshold': 'BLOCK_ONLY_HIGH'}]
|
|
# Initial payload for the first message
|
|
payload: str | list[dict[str, Any]] = user_message
|
|
if adapter.session_id is None:
|
|
if discussion_history:
|
|
payload = f"[DISCUSSION HISTORY]\n\n{discussion_history}\n\n---\n\n{user_message}"
|
|
all_text: list[str] = []
|
|
_cumulative_tool_bytes = 0
|
|
for r_idx in range(MAX_TOOL_ROUNDS + 2):
|
|
if adapter is None:
|
|
break
|
|
events.emit("request_start", payload={"provider": "gemini_cli", "model": _model, "round": r_idx})
|
|
_append_comms("OUT", "request", {"message": f"[CLI] [round {r_idx}] [msg {len(payload)}]"})
|
|
resp_data = adapter.send(payload, safety_settings=safety_settings, system_instruction=sys_instr, model=_model)
|
|
# Log any stderr from the CLI for transparency
|
|
cli_stderr = resp_data.get("stderr", "")
|
|
if cli_stderr:
|
|
sys.stderr.write(f"\n--- Gemini CLI stderr ---\n{cli_stderr}\n-------------------------\n")
|
|
sys.stderr.flush()
|
|
txt = resp_data.get("text", "")
|
|
if txt: all_text.append(txt)
|
|
calls = resp_data.get("tool_calls", [])
|
|
usage = adapter.last_usage or {}
|
|
latency = adapter.last_latency
|
|
events.emit("response_received", payload={"provider": "gemini_cli", "model": _model, "usage": usage, "latency": latency, "round": r_idx})
|
|
# Clean up the tool calls format to match comms log expectation
|
|
log_calls = []
|
|
for c in calls:
|
|
log_calls.append({"name": c.get("name"), "args": c.get("args"), "id": c.get("id")})
|
|
_append_comms("IN", "response", {
|
|
"round": r_idx,
|
|
"stop_reason": "TOOL_USE" if calls else "STOP",
|
|
"text": txt,
|
|
"tool_calls": log_calls,
|
|
"usage": usage
|
|
})
|
|
# If there's text and we're not done, push it to the history immediately
|
|
# so it appears as a separate entry in the GUI.
|
|
if txt and calls and comms_log_callback:
|
|
# Use kind='history_add' to push a new entry into the disc_entries list
|
|
comms_log_callback({
|
|
"ts": project_manager.now_ts(),
|
|
"direction": "IN",
|
|
"kind": "history_add",
|
|
"payload": {
|
|
"role": "AI",
|
|
"content": txt
|
|
}
|
|
})
|
|
if not calls or r_idx > MAX_TOOL_ROUNDS:
|
|
break
|
|
tool_results_for_cli: list[dict[str, Any]] = []
|
|
for i, fc in enumerate(calls):
|
|
name = fc.get("name")
|
|
args = fc.get("args", {})
|
|
call_id = fc.get("id")
|
|
# Check for tool confirmation if callback is provided
|
|
if pre_tool_callback:
|
|
payload_str = json.dumps({"tool": name, "args": args})
|
|
if not pre_tool_callback(payload_str):
|
|
out = "USER REJECTED: tool execution cancelled"
|
|
tool_results_for_cli.append({
|
|
"role": "tool",
|
|
"tool_call_id": call_id,
|
|
"name": name,
|
|
"content": out
|
|
})
|
|
_append_comms("IN", "tool_result", {"name": name, "id": call_id, "output": out})
|
|
continue
|
|
events.emit("tool_execution", payload={"status": "started", "tool": name, "args": args, "round": r_idx})
|
|
if name in mcp_client.TOOL_NAMES:
|
|
_append_comms("OUT", "tool_call", {"name": name, "id": call_id, "args": args})
|
|
out = mcp_client.dispatch(name, args)
|
|
elif name == TOOL_NAME:
|
|
scr = args.get("script", "")
|
|
_append_comms("OUT", "tool_call", {"name": TOOL_NAME, "id": call_id, "script": scr})
|
|
out = _run_script(scr, base_dir, qa_callback)
|
|
else:
|
|
out = f"ERROR: unknown tool '{name}'"
|
|
if i == len(calls) - 1:
|
|
if file_items:
|
|
file_items, changed = _reread_file_items(file_items)
|
|
ctx = _build_file_diff_text(changed)
|
|
if ctx:
|
|
out += f"\n\n[SYSTEM: FILES UPDATED]\n\n{ctx}"
|
|
if r_idx == MAX_TOOL_ROUNDS:
|
|
out += "\n\n[SYSTEM: MAX ROUNDS. PROVIDE FINAL ANSWER.]"
|
|
out = _truncate_tool_output(out)
|
|
_cumulative_tool_bytes += len(out)
|
|
tool_results_for_cli.append({
|
|
"role": "tool",
|
|
"tool_call_id": call_id,
|
|
"name": name,
|
|
"content": out
|
|
})
|
|
_append_comms("IN", "tool_result", {"name": name, "id": call_id, "output": out})
|
|
events.emit("tool_execution", payload={"status": "completed", "tool": name, "result": out, "round": r_idx})
|
|
# CRITICAL: Update payload for the next round
|
|
payload = tool_results_for_cli
|
|
if _cumulative_tool_bytes > _MAX_TOOL_OUTPUT_BYTES:
|
|
_append_comms("OUT", "request", {"message": f"[TOOL OUTPUT BUDGET EXCEEDED: {_cumulative_tool_bytes} bytes]"})
|
|
# We should ideally tell the model here, but for CLI we just append to payload
|
|
# For Gemini CLI, we send the tool results as a JSON array of messages (or similar)
|
|
# The adapter expects a string, so we'll pass the JSON string of the results.
|
|
# Return only the text from the last round, because intermediate
|
|
# text chunks were already pushed to history via comms_log_callback.
|
|
final_text = all_text[-1] if all_text else "(No text returned)"
|
|
return final_text
|
|
except Exception as e:
|
|
# Basic error classification for CLI
|
|
raise ProviderError("unknown", "gemini_cli", e)
|
|
# ------------------------------------------------------------------ anthropic history management
|
|
# Rough chars-per-token ratio. Anthropic tokeniser averages ~3.5-4 chars/token.
|
|
# We use 3.5 to be conservative (overestimate token count = safer).
|
|
_CHARS_PER_TOKEN = 3.5
|
|
|
|
# Maximum token budget for the entire prompt (system + tools + messages).
|
|
# Anthropic's limit is 200k. We leave headroom for the response + tool schemas.
|
|
_ANTHROPIC_MAX_PROMPT_TOKENS = 180_000
|
|
|
|
# Gemini models have a 1M context window but we cap well below to leave headroom.
|
|
# If the model reports input tokens exceeding this, we trim old history.
|
|
_GEMINI_MAX_INPUT_TOKENS = 900_000
|
|
|
|
# Marker prefix used to identify stale file-refresh injections in history
|
|
_FILE_REFRESH_MARKER = "[FILES UPDATED"
|
|
|
|
def _estimate_message_tokens(msg: dict) -> int:
|
|
"""
|
|
Rough token estimate for a single Anthropic message dict.
|
|
Caches the result on the dict as '_est_tokens' so repeated calls
|
|
(e.g., from _trim_anthropic_history) don't re-scan unchanged messages.
|
|
Call _invalidate_token_estimate() when a message's content is modified.
|
|
"""
|
|
cached = msg.get("_est_tokens")
|
|
if cached is not None:
|
|
return cached
|
|
total_chars = 0
|
|
content = msg.get("content", "")
|
|
if isinstance(content, str):
|
|
total_chars += len(content)
|
|
elif isinstance(content, list):
|
|
for block in content:
|
|
if isinstance(block, dict):
|
|
text = block.get("text", "") or block.get("content", "")
|
|
if isinstance(text, str):
|
|
total_chars += len(text)
|
|
# tool_use input
|
|
inp = block.get("input")
|
|
if isinstance(inp, dict):
|
|
import json as _json
|
|
total_chars += len(_json.dumps(inp, ensure_ascii=False))
|
|
elif isinstance(block, str):
|
|
total_chars += len(block)
|
|
est = max(1, int(total_chars / _CHARS_PER_TOKEN))
|
|
msg["_est_tokens"] = est
|
|
return est
|
|
|
|
def _invalidate_token_estimate(msg: dict[str, Any]) -> None:
|
|
"""Remove the cached token estimate so the next call recalculates."""
|
|
msg.pop("_est_tokens", None)
|
|
|
|
def _estimate_prompt_tokens(system_blocks: list[dict], history: list[dict]) -> int:
|
|
"""Estimate total prompt tokens: system + tools + all history messages."""
|
|
total = 0
|
|
# System blocks
|
|
for block in system_blocks:
|
|
text = block.get("text", "")
|
|
total += max(1, int(len(text) / _CHARS_PER_TOKEN))
|
|
# Tool definitions (rough fixed estimate — they're ~2k tokens for our set)
|
|
total += 2500
|
|
# History messages (uses cached estimates for unchanged messages)
|
|
for msg in history:
|
|
total += _estimate_message_tokens(msg)
|
|
return total
|
|
|
|
def _strip_stale_file_refreshes(history: list[dict[str, Any]]) -> None:
|
|
"""
|
|
Remove [FILES UPDATED ...] text blocks from all history turns EXCEPT
|
|
the very last user message. These are stale snapshots from previous
|
|
tool rounds that bloat the context without providing value.
|
|
"""
|
|
if len(history) < 2:
|
|
return
|
|
# Find the index of the last user message — we keep its file refresh intact
|
|
last_user_idx = -1
|
|
for i in range(len(history) - 1, -1, -1):
|
|
if history[i].get("role") == "user":
|
|
last_user_idx = i
|
|
break
|
|
for i, msg in enumerate(history):
|
|
if msg.get("role") != "user" or i == last_user_idx:
|
|
continue
|
|
content = msg.get("content")
|
|
if not isinstance(content, list):
|
|
continue
|
|
cleaned = []
|
|
for block in content:
|
|
if isinstance(block, dict) and block.get("type") == "text":
|
|
text = block.get("text", "")
|
|
if text.startswith(_FILE_REFRESH_MARKER):
|
|
continue # drop this stale file refresh block
|
|
cleaned.append(block)
|
|
if len(cleaned) < len(content):
|
|
msg["content"] = cleaned
|
|
_invalidate_token_estimate(msg)
|
|
|
|
def _trim_anthropic_history(system_blocks: list[dict[str, Any]], history: list[dict[str, Any]]) -> int:
|
|
"""
|
|
Trim the Anthropic history to fit within the token budget.
|
|
Strategy:
|
|
1. Strip stale file-refresh injections from old turns.
|
|
2. If still over budget, drop oldest turn pairs (user + assistant).
|
|
Returns the number of messages dropped.
|
|
"""
|
|
# Phase 1: strip stale file refreshes
|
|
_strip_stale_file_refreshes(history)
|
|
est = _estimate_prompt_tokens(system_blocks, history)
|
|
if est <= _ANTHROPIC_MAX_PROMPT_TOKENS:
|
|
return 0
|
|
# Phase 2: drop oldest turn pairs until within budget
|
|
dropped = 0
|
|
while len(history) > 3 and est > _ANTHROPIC_MAX_PROMPT_TOKENS:
|
|
# Protect history[0] (original user prompt). Drop from history[1] (assistant) and history[2] (user)
|
|
if history[1].get("role") == "assistant" and len(history) > 2 and history[2].get("role") == "user":
|
|
removed_asst = history.pop(1)
|
|
removed_user = history.pop(1)
|
|
dropped += 2
|
|
est -= _estimate_message_tokens(removed_asst)
|
|
est -= _estimate_message_tokens(removed_user)
|
|
# Also drop dangling tool_results if the next message is an assistant and the removed user was just tool results
|
|
while len(history) > 2 and history[1].get("role") == "assistant" and history[2].get("role") == "user":
|
|
content = history[2].get("content", [])
|
|
if isinstance(content, list) and content and isinstance(content[0], dict) and content[0].get("type") == "tool_result":
|
|
r_a = history.pop(1)
|
|
r_u = history.pop(1)
|
|
dropped += 2
|
|
est -= _estimate_message_tokens(r_a)
|
|
est -= _estimate_message_tokens(r_u)
|
|
else:
|
|
break
|
|
else:
|
|
# Edge case fallback: drop index 1 (protecting index 0)
|
|
removed = history.pop(1)
|
|
dropped += 1
|
|
est -= _estimate_message_tokens(removed)
|
|
return dropped
|
|
# ------------------------------------------------------------------ anthropic
|
|
|
|
def _ensure_anthropic_client() -> None:
|
|
global _anthropic_client
|
|
if _anthropic_client is None:
|
|
creds = _load_credentials()
|
|
# Enable prompt caching beta
|
|
_anthropic_client = anthropic.Anthropic(
|
|
api_key=creds["anthropic"]["api_key"],
|
|
default_headers={"anthropic-beta": "prompt-caching-2024-07-31"}
|
|
)
|
|
|
|
def _chunk_text(text: str, chunk_size: int) -> list[str]:
|
|
return [text[i:i + chunk_size] for i in range(0, len(text), chunk_size)]
|
|
|
|
def _build_chunked_context_blocks(md_content: str) -> list[dict]:
|
|
"""
|
|
Split md_content into <=_ANTHROPIC_CHUNK_SIZE char chunks.
|
|
cache_control:ephemeral is placed only on the LAST block so the whole
|
|
prefix is cached as one unit.
|
|
"""
|
|
chunks = _chunk_text(md_content, _ANTHROPIC_CHUNK_SIZE)
|
|
blocks = []
|
|
for i, chunk in enumerate(chunks):
|
|
block: dict = {"type": "text", "text": chunk}
|
|
if i == len(chunks) - 1:
|
|
block["cache_control"] = {"type": "ephemeral"}
|
|
blocks.append(block)
|
|
return blocks
|
|
|
|
def _strip_cache_controls(history: list[dict[str, Any]]) -> None:
|
|
"""
|
|
Remove cache_control from all content blocks in message history.
|
|
Anthropic allows max 4 cache_control blocks total across system + tools +
|
|
messages. We reserve those slots for the stable system/tools prefix and
|
|
the current turn's context block, so all older history entries must be clean.
|
|
"""
|
|
for msg in history:
|
|
content = msg.get("content")
|
|
if isinstance(content, list):
|
|
for block in content:
|
|
if isinstance(block, dict):
|
|
block.pop("cache_control", None)
|
|
|
|
def _add_history_cache_breakpoint(history: list[dict[str, Any]]) -> None:
|
|
"""
|
|
Place cache_control:ephemeral on the last content block of the
|
|
second-to-last user message. This uses one of the 4 allowed Anthropic
|
|
cache breakpoints to cache the conversation prefix so the full history
|
|
isn't reprocessed on every request.
|
|
"""
|
|
user_indices = [i for i, m in enumerate(history) if m.get("role") == "user"]
|
|
if len(user_indices) < 2:
|
|
return # Only one user message (the current turn) — nothing stable to cache
|
|
target_idx = user_indices[-2]
|
|
content = history[target_idx].get("content")
|
|
if isinstance(content, list) and content:
|
|
last_block = content[-1]
|
|
if isinstance(last_block, dict):
|
|
last_block["cache_control"] = {"type": "ephemeral"}
|
|
elif isinstance(content, str):
|
|
history[target_idx]["content"] = [
|
|
{"type": "text", "text": content, "cache_control": {"type": "ephemeral"}}
|
|
]
|
|
|
|
def _repair_anthropic_history(history: list[dict[str, Any]]) -> None:
|
|
"""
|
|
If history ends with an assistant message that contains tool_use blocks
|
|
without a following user tool_result message, append a synthetic tool_result
|
|
message so the history is valid before the next request.
|
|
"""
|
|
if not history:
|
|
return
|
|
last = history[-1]
|
|
if last.get("role") != "assistant":
|
|
return
|
|
content = last.get("content", [])
|
|
tool_use_ids = []
|
|
for block in content:
|
|
if isinstance(block, dict):
|
|
if block.get("type") == "tool_use":
|
|
tool_use_ids.append(block["id"])
|
|
if not tool_use_ids:
|
|
return
|
|
history.append({
|
|
"role": "user",
|
|
"content": [
|
|
{
|
|
"type": "tool_result",
|
|
"tool_use_id": tid,
|
|
"content": "Tool call was not completed (session interrupted).",
|
|
}
|
|
for tid in tool_use_ids
|
|
],
|
|
})
|
|
|
|
def _send_anthropic(md_content: str, user_message: str, base_dir: str, file_items: list[dict[str, Any]] | None = None, discussion_history: str = "", pre_tool_callback: Optional[Callable[[str], bool]] = None, qa_callback: Optional[Callable[[str], str]] = None) -> str:
|
|
try:
|
|
_ensure_anthropic_client()
|
|
mcp_client.configure(file_items or [], [base_dir])
|
|
# Split system into two cache breakpoints:
|
|
# 1. Stable system prompt (never changes — always a cache hit)
|
|
# 2. Dynamic file context (invalidated only when files change)
|
|
stable_prompt = _get_combined_system_prompt()
|
|
stable_blocks = [{"type": "text", "text": stable_prompt, "cache_control": {"type": "ephemeral"}}]
|
|
context_text = f"\n\n<context>\n{md_content}\n</context>"
|
|
context_blocks = _build_chunked_context_blocks(context_text)
|
|
system_blocks = stable_blocks + context_blocks
|
|
# Prepend discussion history to the first user message if this is a fresh session
|
|
if discussion_history and not _anthropic_history:
|
|
user_content: list[dict[str, Any]] = [{"type": "text", "text": f"[DISCUSSION HISTORY]\n\n{discussion_history}\n\n---\n\n{user_message}"}]
|
|
else:
|
|
user_content = [{"type": "text", "text": user_message}]
|
|
# COMPRESS HISTORY: Truncate massive tool outputs from previous turns
|
|
for msg in _anthropic_history:
|
|
if msg.get("role") == "user" and isinstance(msg.get("content"), list):
|
|
modified = False
|
|
for block in msg["content"]:
|
|
if isinstance(block, dict) and block.get("type") == "tool_result":
|
|
t_content = block.get("content", "")
|
|
if _history_trunc_limit > 0 and isinstance(t_content, str) and len(t_content) > _history_trunc_limit:
|
|
block["content"] = t_content[:_history_trunc_limit] + "\n\n... [TRUNCATED BY SYSTEM TO SAVE TOKENS. Original output was too large.]"
|
|
modified = True
|
|
if modified:
|
|
_invalidate_token_estimate(msg)
|
|
_strip_cache_controls(_anthropic_history)
|
|
_repair_anthropic_history(_anthropic_history)
|
|
_anthropic_history.append({"role": "user", "content": user_content})
|
|
# Use the 4th cache breakpoint to cache the conversation history prefix.
|
|
# This is placed on the second-to-last user message (the last stable one).
|
|
_add_history_cache_breakpoint(_anthropic_history)
|
|
n_chunks = len(system_blocks)
|
|
_append_comms("OUT", "request", {
|
|
"message": (
|
|
f"[system {n_chunks} chunk(s), {len(md_content)} chars context] "
|
|
f"{user_message[:200]}{'...' if len(user_message) > 200 else ''}"
|
|
),
|
|
})
|
|
all_text_parts: list[str] = []
|
|
_cumulative_tool_bytes = 0
|
|
# We allow MAX_TOOL_ROUNDS, plus 1 final loop to get the text synthesis
|
|
for round_idx in range(MAX_TOOL_ROUNDS + 2):
|
|
# Trim history to fit within token budget before each API call
|
|
dropped = _trim_anthropic_history(system_blocks, _anthropic_history)
|
|
if dropped > 0:
|
|
est_tokens = _estimate_prompt_tokens(system_blocks, _anthropic_history)
|
|
_append_comms("OUT", "request", {
|
|
"message": (
|
|
f"[HISTORY TRIMMED: dropped {dropped} old messages to fit token budget. "
|
|
f"Estimated {est_tokens} tokens remaining. {len(_anthropic_history)} messages in history.]"
|
|
),
|
|
})
|
|
|
|
def _strip_private_keys(history: list[dict[str, Any]]) -> list[dict[str, Any]]:
|
|
return [{k: v for k, v in m.items() if not k.startswith("_")} for m in history]
|
|
events.emit("request_start", payload={"provider": "anthropic", "model": _model, "round": round_idx})
|
|
response = _anthropic_client.messages.create(
|
|
model=_model,
|
|
max_tokens=_max_tokens,
|
|
temperature=_temperature,
|
|
system=system_blocks,
|
|
tools=_get_anthropic_tools(),
|
|
messages=_strip_private_keys(_anthropic_history),
|
|
)
|
|
# Convert SDK content block objects to plain dicts before storing in history
|
|
serialised_content = [_content_block_to_dict(b) for b in response.content]
|
|
_anthropic_history.append({
|
|
"role": "assistant",
|
|
"content": serialised_content,
|
|
})
|
|
text_blocks = [b.text for b in response.content if hasattr(b, "text") and b.text]
|
|
if text_blocks:
|
|
all_text_parts.append("\n".join(text_blocks))
|
|
tool_use_blocks = [
|
|
{"id": b.id, "name": b.name, "input": b.input}
|
|
for b in response.content
|
|
if getattr(b, "type", None) == "tool_use"
|
|
]
|
|
usage_dict: dict[str, Any] = {}
|
|
if response.usage:
|
|
usage_dict["input_tokens"] = response.usage.input_tokens
|
|
usage_dict["output_tokens"] = response.usage.output_tokens
|
|
cache_creation = getattr(response.usage, "cache_creation_input_tokens", None)
|
|
cache_read = getattr(response.usage, "cache_read_input_tokens", None)
|
|
if cache_creation is not None:
|
|
usage_dict["cache_creation_input_tokens"] = cache_creation
|
|
if cache_read is not None:
|
|
usage_dict["cache_read_input_tokens"] = cache_read
|
|
events.emit("response_received", payload={"provider": "anthropic", "model": _model, "usage": usage_dict, "round": round_idx})
|
|
_append_comms("IN", "response", {
|
|
"round": round_idx,
|
|
"stop_reason": response.stop_reason,
|
|
"text": "\n".join(text_blocks),
|
|
"tool_calls": tool_use_blocks,
|
|
"usage": usage_dict,
|
|
})
|
|
if response.stop_reason != "tool_use" or not tool_use_blocks:
|
|
break
|
|
if round_idx > MAX_TOOL_ROUNDS:
|
|
# The model ignored the MAX ROUNDS warning and kept calling tools.
|
|
# Force abort to prevent infinite loop.
|
|
break
|
|
tool_results: list[dict[str, Any]] = []
|
|
for block in response.content:
|
|
if getattr(block, "type", None) != "tool_use":
|
|
continue
|
|
b_name = getattr(block, "name", None)
|
|
b_id = getattr(block, "id", "")
|
|
b_input = getattr(block, "input", {})
|
|
# Check for tool confirmation if callback is provided
|
|
if pre_tool_callback:
|
|
payload_str = json.dumps({"tool": b_name, "args": b_input})
|
|
if not pre_tool_callback(payload_str):
|
|
output = "USER REJECTED: tool execution cancelled"
|
|
tool_results.append({
|
|
"type": "tool_result",
|
|
"tool_use_id": b_id,
|
|
"content": output,
|
|
})
|
|
continue
|
|
events.emit("tool_execution", payload={"status": "started", "tool": b_name, "args": b_input, "round": round_idx})
|
|
if b_name in mcp_client.TOOL_NAMES:
|
|
_append_comms("OUT", "tool_call", {"name": b_name, "id": b_id, "args": b_input})
|
|
output = mcp_client.dispatch(b_name, b_input)
|
|
_append_comms("IN", "tool_result", {"name": b_name, "id": b_id, "output": output})
|
|
truncated = _truncate_tool_output(output)
|
|
_cumulative_tool_bytes += len(truncated)
|
|
tool_results.append({
|
|
"type": "tool_result",
|
|
"tool_use_id": b_id,
|
|
"content": truncated,
|
|
})
|
|
events.emit("tool_execution", payload={"status": "completed", "tool": b_name, "result": output, "round": round_idx})
|
|
elif b_name == TOOL_NAME:
|
|
script = b_input.get("script", "")
|
|
_append_comms("OUT", "tool_call", {
|
|
"name": TOOL_NAME,
|
|
"id": b_id,
|
|
"script": script,
|
|
})
|
|
output = _run_script(script, base_dir, qa_callback)
|
|
_append_comms("IN", "tool_result", {
|
|
"name": TOOL_NAME,
|
|
"id": b_id,
|
|
"output": output,
|
|
})
|
|
truncated = _truncate_tool_output(output)
|
|
_cumulative_tool_bytes += len(truncated)
|
|
tool_results.append({
|
|
"type": "tool_result",
|
|
"tool_use_id": b_id,
|
|
"content": truncated,
|
|
})
|
|
events.emit("tool_execution", payload={"status": "completed", "tool": b_name, "result": output, "round": round_idx})
|
|
if _cumulative_tool_bytes > _MAX_TOOL_OUTPUT_BYTES:
|
|
tool_results.append({
|
|
"type": "text",
|
|
"text": f"SYSTEM WARNING: Cumulative tool output exceeded {_MAX_TOOL_OUTPUT_BYTES // 1000}KB budget. Provide your final answer now."
|
|
})
|
|
_append_comms("OUT", "request", {"message": f"[TOOL OUTPUT BUDGET EXCEEDED: {_cumulative_tool_bytes} bytes]"})
|
|
# Refresh file context after tool calls — only inject CHANGED files
|
|
if file_items:
|
|
file_items, changed = _reread_file_items(file_items)
|
|
refreshed_ctx = _build_file_diff_text(changed)
|
|
if refreshed_ctx:
|
|
tool_results.append({
|
|
"type": "text",
|
|
"text": (
|
|
"[FILES UPDATED — current contents below. "
|
|
"Do NOT re-read these files with PowerShell.]\n\n"
|
|
+ refreshed_ctx
|
|
),
|
|
})
|
|
if round_idx == MAX_TOOL_ROUNDS:
|
|
tool_results.append({
|
|
"type": "text",
|
|
"text": "SYSTEM WARNING: MAX TOOL ROUNDS REACHED. YOU MUST PROVIDE YOUR FINAL ANSWER NOW WITHOUT CALLING ANY MORE TOOLS."
|
|
})
|
|
_anthropic_history.append({
|
|
"role": "user",
|
|
"content": tool_results,
|
|
})
|
|
_append_comms("OUT", "tool_result_send", {
|
|
"results": [
|
|
{"tool_use_id": r["tool_use_id"], "content": r["content"]}
|
|
for r in tool_results if r.get("type") == "tool_result"
|
|
],
|
|
})
|
|
final_text = "\n\n".join(all_text_parts)
|
|
return final_text if final_text.strip() else "(No text returned by the model)"
|
|
except ProviderError:
|
|
raise
|
|
except Exception as exc:
|
|
raise _classify_anthropic_error(exc) from exc
|
|
# ------------------------------------------------------------------ deepseek
|
|
|
|
def _ensure_deepseek_client() -> None:
|
|
global _deepseek_client
|
|
if _deepseek_client is None:
|
|
creds = _load_credentials()
|
|
# Placeholder for Dedicated DeepSeek SDK instantiation
|
|
# import deepseek
|
|
# _deepseek_client = deepseek.DeepSeek(api_key=creds["deepseek"]["api_key"])
|
|
pass
|
|
|
|
def _send_deepseek(md_content: str, user_message: str, base_dir: str,
|
|
file_items: list[dict[str, Any]] | None = None,
|
|
discussion_history: str = "",
|
|
stream: bool = False,
|
|
pre_tool_callback: Optional[Callable[[str], bool]] = None,
|
|
qa_callback: Optional[Callable[[str], str]] = None) -> str:
|
|
"""
|
|
Sends a message to the DeepSeek API, handling tool calls and history.
|
|
Supports streaming responses.
|
|
"""
|
|
try:
|
|
mcp_client.configure(file_items or [], [base_dir])
|
|
creds = _load_credentials()
|
|
api_key = creds.get("deepseek", {}).get("api_key")
|
|
if not api_key:
|
|
raise ValueError("DeepSeek API key not found in credentials.toml")
|
|
# DeepSeek API details
|
|
api_url = "https://api.deepseek.com/chat/completions"
|
|
headers = {
|
|
"Authorization": f"Bearer {api_key}",
|
|
"Content-Type": "application/json",
|
|
}
|
|
# Build the messages for the current API call
|
|
current_api_messages: list[dict[str, Any]] = []
|
|
with _deepseek_history_lock:
|
|
for msg in _deepseek_history:
|
|
current_api_messages.append(msg)
|
|
# Add the current user's input for this turn
|
|
initial_user_message_content = user_message
|
|
if discussion_history:
|
|
initial_user_message_content = f"[DISCUSSION HISTORY]\n\n{discussion_history}\n\n---\n\n{user_message}"
|
|
current_api_messages.append({"role": "user", "content": initial_user_message_content})
|
|
# Construct the full request payload
|
|
request_payload: dict[str, Any] = {
|
|
"model": _model,
|
|
"messages": current_api_messages,
|
|
"temperature": _temperature,
|
|
"max_tokens": _max_tokens,
|
|
"stream": stream,
|
|
}
|
|
# Insert system prompt at the beginning
|
|
sys_msg = {"role": "system", "content": f"{_get_combined_system_prompt()}\n\n<context>\n{md_content}\n</context>"}
|
|
request_payload["messages"].insert(0, sys_msg)
|
|
all_text_parts: list[str] = []
|
|
_cumulative_tool_bytes = 0
|
|
round_idx = 0
|
|
while round_idx <= MAX_TOOL_ROUNDS + 1:
|
|
events.emit("request_start", payload={"provider": "deepseek", "model": _model, "round": round_idx, "streaming": stream})
|
|
try:
|
|
response = requests.post(api_url, headers=headers, json=request_payload, timeout=60, stream=stream)
|
|
response.raise_for_status()
|
|
except requests.exceptions.RequestException as e:
|
|
raise _classify_deepseek_error(e) from e
|
|
# Process response
|
|
if stream:
|
|
aggregated_content = ""
|
|
aggregated_tool_calls: list[dict[str, Any]] = []
|
|
aggregated_reasoning = ""
|
|
current_usage: dict[str, Any] = {}
|
|
final_finish_reason = "stop"
|
|
for line in response.iter_lines():
|
|
if not line:
|
|
continue
|
|
decoded = line.decode('utf-8')
|
|
if decoded.startswith('data: '):
|
|
chunk_str = decoded[len('data: '):]
|
|
if chunk_str.strip() == '[DONE]':
|
|
continue
|
|
try:
|
|
chunk = json.loads(chunk_str)
|
|
delta = chunk.get("choices", [{}])[0].get("delta", {})
|
|
if delta.get("content"):
|
|
aggregated_content += delta["content"]
|
|
if delta.get("reasoning_content"):
|
|
aggregated_reasoning += delta["reasoning_content"]
|
|
if delta.get("tool_calls"):
|
|
# Simple aggregation of tool call deltas
|
|
for tc_delta in delta["tool_calls"]:
|
|
idx = tc_delta.get("index", 0)
|
|
while len(aggregated_tool_calls) <= idx:
|
|
aggregated_tool_calls.append({"id": "", "type": "function", "function": {"name": "", "arguments": ""}})
|
|
target = aggregated_tool_calls[idx]
|
|
if tc_delta.get("id"):
|
|
target["id"] = tc_delta["id"]
|
|
if tc_delta.get("function", {}).get("name"):
|
|
target["function"]["name"] += tc_delta["function"]["name"]
|
|
if tc_delta.get("function", {}).get("arguments"):
|
|
target["function"]["arguments"] += tc_delta["function"]["arguments"]
|
|
if chunk.get("choices", [{}])[0].get("finish_reason"):
|
|
final_finish_reason = chunk["choices"][0]["finish_reason"]
|
|
if chunk.get("usage"):
|
|
current_usage = chunk["usage"]
|
|
except json.JSONDecodeError:
|
|
continue
|
|
assistant_text = aggregated_content
|
|
tool_calls_raw = aggregated_tool_calls
|
|
reasoning_content = aggregated_reasoning
|
|
finish_reason = final_finish_reason
|
|
usage = current_usage
|
|
else:
|
|
response_data = response.json()
|
|
choices = response_data.get("choices", [])
|
|
if not choices:
|
|
_append_comms("IN", "response", {"round": round_idx, "text": "(No choices returned)", "usage": response_data.get("usage", {})})
|
|
break
|
|
choice = choices[0]
|
|
message = choice.get("message", {})
|
|
assistant_text = message.get("content", "")
|
|
tool_calls_raw = message.get("tool_calls", [])
|
|
reasoning_content = message.get("reasoning_content", "")
|
|
finish_reason = choice.get("finish_reason", "stop")
|
|
usage = response_data.get("usage", {})
|
|
# Format reasoning content if it exists
|
|
thinking_tags = ""
|
|
if reasoning_content:
|
|
thinking_tags = f"<thinking>\n{reasoning_content}\n</thinking>\n"
|
|
full_assistant_text = thinking_tags + assistant_text
|
|
# Update history
|
|
with _deepseek_history_lock:
|
|
msg_to_store: dict[str, Any] = {"role": "assistant", "content": assistant_text}
|
|
if reasoning_content:
|
|
msg_to_store["reasoning_content"] = reasoning_content
|
|
if tool_calls_raw:
|
|
msg_to_store["tool_calls"] = tool_calls_raw
|
|
_deepseek_history.append(msg_to_store)
|
|
if full_assistant_text:
|
|
all_text_parts.append(full_assistant_text)
|
|
_append_comms("IN", "response", {
|
|
"round": round_idx,
|
|
"stop_reason": finish_reason,
|
|
"text": full_assistant_text,
|
|
"tool_calls": tool_calls_raw,
|
|
"usage": usage,
|
|
"streaming": stream
|
|
})
|
|
if finish_reason != "tool_calls" and not tool_calls_raw:
|
|
break
|
|
if round_idx > MAX_TOOL_ROUNDS:
|
|
break
|
|
tool_results_for_history: list[dict[str, Any]] = []
|
|
for i, tc_raw in enumerate(tool_calls_raw):
|
|
tool_info = tc_raw.get("function", {})
|
|
tool_name = tool_info.get("name")
|
|
tool_args_str = tool_info.get("arguments", "{}")
|
|
tool_id = tc_raw.get("id")
|
|
try:
|
|
tool_args = json.loads(tool_args_str)
|
|
except:
|
|
tool_args = {}
|
|
# Check for tool confirmation if callback is provided
|
|
if pre_tool_callback:
|
|
payload_str = json.dumps({"tool": tool_name, "args": tool_args})
|
|
if not pre_tool_callback(payload_str):
|
|
tool_output = "USER REJECTED: tool execution cancelled"
|
|
tool_results_for_history.append({
|
|
"role": "tool",
|
|
"tool_call_id": tool_id,
|
|
"content": tool_output,
|
|
})
|
|
_append_comms("IN", "tool_result", {"name": tool_name, "id": tool_id, "output": tool_output})
|
|
continue
|
|
events.emit("tool_execution", payload={"status": "started", "tool": tool_name, "args": tool_args, "round": round_idx})
|
|
if tool_name in mcp_client.TOOL_NAMES:
|
|
_append_comms("OUT", "tool_call", {"name": tool_name, "id": tool_id, "args": tool_args})
|
|
tool_output = mcp_client.dispatch(tool_name, tool_args)
|
|
elif tool_name == TOOL_NAME:
|
|
script = tool_args.get("script", "")
|
|
_append_comms("OUT", "tool_call", {"name": TOOL_NAME, "id": tool_id, "script": script})
|
|
tool_output = _run_script(script, base_dir, qa_callback)
|
|
else:
|
|
tool_output = f"ERROR: unknown tool '{tool_name}'"
|
|
if i == len(tool_calls_raw) - 1:
|
|
if file_items:
|
|
file_items, changed = _reread_file_items(file_items)
|
|
ctx = _build_file_diff_text(changed)
|
|
if ctx:
|
|
tool_output += f"\n\n[SYSTEM: FILES UPDATED]\n\n{ctx}"
|
|
if round_idx == MAX_TOOL_ROUNDS:
|
|
tool_output += "\n\n[SYSTEM: MAX ROUNDS. PROVIDE FINAL ANSWER.]"
|
|
tool_output = _truncate_tool_output(tool_output)
|
|
_cumulative_tool_bytes += len(tool_output)
|
|
tool_results_for_history.append({
|
|
"role": "tool",
|
|
"tool_call_id": tool_id,
|
|
"content": tool_output,
|
|
})
|
|
_append_comms("IN", "tool_result", {"name": tool_name, "id": tool_id, "output": tool_output})
|
|
events.emit("tool_execution", payload={"status": "completed", "tool": tool_name, "result": tool_output, "round": round_idx})
|
|
if _cumulative_tool_bytes > _MAX_TOOL_OUTPUT_BYTES:
|
|
tool_results_for_history.append({
|
|
"role": "user",
|
|
"content": f"SYSTEM WARNING: Cumulative tool output exceeded {_MAX_TOOL_OUTPUT_BYTES // 1000}KB budget. Provide your final answer now."
|
|
})
|
|
_append_comms("OUT", "request", {"message": f"[TOOL OUTPUT BUDGET EXCEEDED: {_cumulative_tool_bytes} bytes]"})
|
|
with _deepseek_history_lock:
|
|
for tr in tool_results_for_history:
|
|
_deepseek_history.append(tr)
|
|
# Update for next round
|
|
next_messages: list[dict[str, Any]] = []
|
|
with _deepseek_history_lock:
|
|
for msg in _deepseek_history:
|
|
next_messages.append(msg)
|
|
next_messages.insert(0, sys_msg)
|
|
request_payload["messages"] = next_messages
|
|
round_idx += 1
|
|
return "\n\n".join(all_text_parts) if all_text_parts else "(No text returned)"
|
|
except Exception as e:
|
|
raise _classify_deepseek_error(e) from e
|
|
|
|
def run_tier4_analysis(stderr: str) -> str:
|
|
"""
|
|
Stateless Tier 4 QA analysis of an error message.
|
|
Uses gemini-2.5-flash-lite to summarize the error and suggest a fix.
|
|
"""
|
|
if not stderr or not stderr.strip():
|
|
return ""
|
|
try:
|
|
_ensure_gemini_client()
|
|
prompt = (
|
|
f"You are a Tier 4 QA Agent specializing in error analysis.\n"
|
|
f"Analyze the following stderr output from a PowerShell command:\n\n"
|
|
f"```\n{stderr}\n```\n\n"
|
|
f"Provide a concise summary of the failure and suggest a fix in approximately 20 words."
|
|
)
|
|
# Use flash-lite for cost-effective stateless analysis
|
|
model_name = "gemini-2.5-flash-lite"
|
|
# We don't use the chat session here to keep it stateless
|
|
resp = _gemini_client.models.generate_content(
|
|
model=model_name,
|
|
contents=prompt,
|
|
config=types.GenerateContentConfig(
|
|
temperature=0.0,
|
|
max_output_tokens=150,
|
|
)
|
|
)
|
|
analysis = resp.text.strip()
|
|
return analysis
|
|
except Exception as e:
|
|
# We don't want to crash the main loop if QA analysis fails
|
|
return f"[QA ANALYSIS FAILED] {e}"
|
|
# ------------------------------------------------------------------ unified send
|
|
|
|
def send(
|
|
md_content: str,
|
|
user_message: str,
|
|
base_dir: str = ".",
|
|
file_items: list[dict[str, Any]] | None = None,
|
|
discussion_history: str = "",
|
|
stream: bool = False,
|
|
pre_tool_callback: Optional[Callable[[str], bool]] = None,
|
|
qa_callback: Optional[Callable[[str], str]] = None,
|
|
) -> str:
|
|
"""
|
|
Send a message to the active provider.
|
|
|
|
md_content : aggregated markdown string (for Gemini: stable content only,
|
|
for Anthropic: full content including history)
|
|
user_message : the user question / instruction
|
|
base_dir : project base directory (for PowerShell tool calls)
|
|
file_items : list of file dicts from aggregate.build_file_items() for
|
|
dynamic context refresh after tool calls
|
|
discussion_history : discussion history text (used by Gemini to inject as
|
|
conversation message instead of caching it)
|
|
stream : Whether to use streaming (supported by DeepSeek)
|
|
pre_tool_callback : Optional callback (payload: str) -> bool called before tool execution
|
|
qa_callback : Optional callback (stderr: str) -> str called for Tier 4 error analysis
|
|
"""
|
|
with _send_lock:
|
|
if _provider == "gemini":
|
|
return _send_gemini(md_content, user_message, base_dir, file_items, discussion_history, pre_tool_callback, qa_callback)
|
|
elif _provider == "gemini_cli":
|
|
return _send_gemini_cli(md_content, user_message, base_dir, file_items, discussion_history, pre_tool_callback, qa_callback)
|
|
elif _provider == "anthropic":
|
|
return _send_anthropic(md_content, user_message, base_dir, file_items, discussion_history, pre_tool_callback, qa_callback)
|
|
elif _provider == "deepseek":
|
|
return _send_deepseek(md_content, user_message, base_dir, file_items, discussion_history, stream=stream, pre_tool_callback=pre_tool_callback, qa_callback=qa_callback)
|
|
raise ValueError(f"unknown provider: {_provider}")
|
|
|
|
def get_history_bleed_stats(md_content: str | None = None) -> dict[str, Any]:
|
|
"""
|
|
Calculates how close the current conversation history is to the token limit.
|
|
If md_content is provided and no chat session exists, it estimates based on md_content.
|
|
"""
|
|
if _provider == "anthropic":
|
|
# For Anthropic, we have a robust estimator
|
|
with _anthropic_history_lock:
|
|
history_snapshot = list(_anthropic_history)
|
|
current_tokens = _estimate_prompt_tokens([], history_snapshot)
|
|
if md_content:
|
|
current_tokens += max(1, int(len(md_content) / _CHARS_PER_TOKEN))
|
|
limit_tokens = _ANTHROPIC_MAX_PROMPT_TOKENS
|
|
percentage = (current_tokens / limit_tokens) * 100 if limit_tokens > 0 else 0
|
|
return {
|
|
"provider": "anthropic",
|
|
"limit": limit_tokens,
|
|
"current": current_tokens,
|
|
"percentage": percentage,
|
|
}
|
|
elif _provider == "gemini":
|
|
effective_limit = _history_trunc_limit if _history_trunc_limit > 0 else _GEMINI_MAX_INPUT_TOKENS
|
|
if _gemini_chat:
|
|
try:
|
|
_ensure_gemini_client()
|
|
raw_history = list(_get_gemini_history_list(_gemini_chat))
|
|
# Copy and correct roles for counting
|
|
history = []
|
|
for c in raw_history:
|
|
# Gemini roles MUST be 'user' or 'model'
|
|
role = "model" if c.role in ["assistant", "model"] else "user"
|
|
history.append(types.Content(role=role, parts=c.parts))
|
|
if md_content:
|
|
# Prepend context as a user part for counting
|
|
history.insert(0, types.Content(role="user", parts=[types.Part.from_text(text=md_content)]))
|
|
if not history:
|
|
print("[DEBUG] Gemini count_tokens skipped: no history or md_content")
|
|
return {
|
|
"provider": "gemini",
|
|
"limit": effective_limit,
|
|
"current": 0,
|
|
"percentage": 0,
|
|
}
|
|
print(f"[DEBUG] Gemini count_tokens on {len(history)} messages using model {_model}")
|
|
resp = _gemini_client.models.count_tokens(
|
|
model=_model,
|
|
contents=history
|
|
)
|
|
current_tokens = resp.total_tokens
|
|
percentage = (current_tokens / effective_limit) * 100 if effective_limit > 0 else 0
|
|
print(f"[DEBUG] Gemini current_tokens={current_tokens}, percentage={percentage:.4f}%")
|
|
return {
|
|
"provider": "gemini",
|
|
"limit": effective_limit,
|
|
"current": current_tokens,
|
|
"percentage": percentage,
|
|
}
|
|
except Exception as e:
|
|
print(f"[DEBUG] Gemini count_tokens error: {e}")
|
|
pass
|
|
elif md_content:
|
|
try:
|
|
_ensure_gemini_client()
|
|
print(f"[DEBUG] Gemini count_tokens (MD ONLY) using model {_model}")
|
|
resp = _gemini_client.models.count_tokens(
|
|
model=_model,
|
|
contents=[types.Content(role="user", parts=[types.Part.from_text(text=md_content)])]
|
|
)
|
|
current_tokens = resp.total_tokens
|
|
percentage = (current_tokens / effective_limit) * 100 if effective_limit > 0 else 0
|
|
print(f"[DEBUG] Gemini (MD ONLY) current_tokens={current_tokens}, percentage={percentage:.4f}%")
|
|
return {
|
|
"provider": "gemini",
|
|
"limit": effective_limit,
|
|
"current": current_tokens,
|
|
"percentage": percentage,
|
|
}
|
|
except Exception as e:
|
|
print(f"[DEBUG] Gemini count_tokens (MD ONLY) error: {e}")
|
|
pass
|
|
return {
|
|
"provider": "gemini",
|
|
"limit": effective_limit,
|
|
"current": 0,
|
|
"percentage": 0,
|
|
}
|
|
elif _provider == "gemini_cli":
|
|
effective_limit = _history_trunc_limit if _history_trunc_limit > 0 else _GEMINI_MAX_INPUT_TOKENS
|
|
# For Gemini CLI, we don't have direct count_tokens access without making a call,
|
|
# so we report the limit and current usage from the last run if available.
|
|
limit_tokens = effective_limit
|
|
current_tokens = 0
|
|
if _gemini_cli_adapter and _gemini_cli_adapter.last_usage:
|
|
# Stats from CLI use 'input_tokens' or 'input'
|
|
u = _gemini_cli_adapter.last_usage
|
|
current_tokens = u.get("input_tokens") or u.get("input", 0)
|
|
percentage = (current_tokens / limit_tokens) * 100 if limit_tokens > 0 else 0
|
|
return {
|
|
"provider": "gemini_cli",
|
|
"limit": limit_tokens,
|
|
"current": current_tokens,
|
|
"percentage": percentage,
|
|
}
|
|
elif _provider == "deepseek":
|
|
limit_tokens = 64000
|
|
current_tokens = 0
|
|
with _deepseek_history_lock:
|
|
for msg in _deepseek_history:
|
|
content = msg.get("content", "")
|
|
if isinstance(content, str):
|
|
current_tokens += len(content)
|
|
elif isinstance(content, list):
|
|
for block in content:
|
|
if isinstance(block, dict):
|
|
text = block.get("text", "")
|
|
if isinstance(text, str):
|
|
current_tokens += len(text)
|
|
inp = block.get("input")
|
|
if isinstance(inp, dict):
|
|
import json as _json
|
|
current_tokens += len(_json.dumps(inp, ensure_ascii=False))
|
|
if md_content:
|
|
current_tokens += len(md_content)
|
|
if user_message:
|
|
current_tokens += len(user_message)
|
|
current_tokens = max(1, int(current_tokens / _CHARS_PER_TOKEN))
|
|
percentage = (current_tokens / limit_tokens) * 100 if limit_tokens > 0 else 0
|
|
return {
|
|
"provider": "deepseek",
|
|
"limit": limit_tokens,
|
|
"current": current_tokens,
|
|
"percentage": percentage,
|
|
}
|
|
# Default empty state
|
|
return {
|
|
"provider": _provider,
|
|
"limit": 0,
|
|
"current": 0,
|
|
"percentage": 0,
|
|
}
|