jpmschweitzerandClaude Opus 4.5
f4e8552298
Implements the first Claude-like agent for codebase exploration: Core Features: - Explore agent with glob, grep, read, and bash tools - Native PydanticAI tool calling with Ollama/Mistral Nemo - Sanitized Ollama provider (fixes content:null issue) - REST API endpoints for agent execution Tool Infrastructure: - BaseTool abstract class with ToolResult dataclass - ReadFileTool, GlobFilesTool, GrepContentTool, BashReadOnlyTool - Path validation and sandboxing support CLI Client (separate package for future extraction): - webber-cli command with chat, explore, status commands - Communicates with Webber API backend - Rich console output with theming Configuration: - Dev server on port 8095 (production uses 8086) - Mistral Nemo optimizations (temp 0.3, tool_choice required) Tests: 24 tests covering tools and API endpoints Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
15 lines
708 B
Markdown
15 lines
708 B
Markdown
## Background
|
|
|
|
Research with Gemini identified key issues with mistral-nemo and tool calling:
|
|
- "Pre-computation Hallucination" - model answers before using tools
|
|
- High default temperature (0.7-0.8) causes wandering
|
|
- Model is "chatty and confident" - needs explicit constraints
|
|
|
|
## Key Recommendations from Gemini Research
|
|
|
|
1. **Temperature 0.0** for tool-calling agents (deterministic, follows schema)
|
|
2. **Chain of Thought (CoT)** - force step-by-step reasoning
|
|
3. **Negative constraints** - tell model what NOT to do (Nemo responds better)
|
|
4. **Explicit tool descriptions** - verbose docstrings with "never estimate yourself"
|
|
5. **"Strictly tool-based assistant"** pattern - NO internal knowledge claim
|