Core orchestrator for self-sustaining AI agent: - Dispatcher: talks to Claude Code CLI with model tier selection - Router: classifies tasks via Haiku, routes to cheapest capable model - Scheduler: queue management + systemd self-wake timers - API: FastAPI endpoints for task execution and monitoring - Ledger: JSONL cost tracking for every inference call Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
25 lines
833 B
Markdown
25 lines
833 B
Markdown
# Symbiont
|
|
|
|
A self-sustaining AI orchestrator that manages its own compute costs through intelligent LLM routing.
|
|
|
|
## Architecture
|
|
|
|
- **Router**: Classifies tasks and dispatches to the cheapest capable model
|
|
- **Dispatcher**: Executes tasks via Claude Code CLI (Pro account) with fallback tiers
|
|
- **Ledger**: Tracks every inference call, cost, and associated revenue
|
|
- **Scheduler**: Handles rate-limit backoff and self-waking timers
|
|
|
|
## Model Tiers
|
|
|
|
| Tier | Model | Use Case |
|
|
|------|-------|----------|
|
|
| 1 | Haiku | Classification, extraction, simple formatting |
|
|
| 2 | Sonnet | Content writing, code gen, analysis |
|
|
| 3 | Opus | Strategic decisions, complex reasoning, QA |
|
|
|
|
## Setup
|
|
|
|
1. Authenticate Claude Code CLI: `claude login`
|
|
2. Install dependencies: `pip install -r requirements.txt`
|
|
3. Run: `python3 -m symbiont.main`
|