The Brain — AI Orchestration for Claude

How it works

Every AI task gets routed to the cheapest capable provider. Claude only handles what only Claude can do.

You → Orchestrator (Claude) → Router
                                    ↓
┌────────────────────────────────────────────────────┐
│ classification / yes-no → Cerebras   (FREE, ~284ms) │
│ factual Q&A / general    → Groq       (FREE, ~366ms) │
│ summarisation / long text→ Gemini     (FREE, 1M ctx) │
│ code generation          → Mistral    (FREE) │
│ translation              → Gemini     (FREE) │
│ deep reasoning / analysis→ SambaNova (FREE, 70B) │
│ creative writing         → Mistral    (FREE) │
│ image generation         → Pollinations (no key) │
└────────────────────────────────────────────────────┘
                                    ↓
                    stats/usage.json (every call logged)

Why it exists

💸

Claude tokens are expensive

Simple tasks like classification, summarisation, and Q&A don't need Claude's intelligence — route them to free providers instead.

🎯

Different AIs excel at different tasks

Cerebras is the fastest. Gemini handles million-token contexts. SambaNova runs the biggest open models. Use each for what it does best.

📋

Every call is logged

All usage goes to stats/usage.json. A nightly report shows how many tokens were saved vs. the cost of running everything through Claude.

🔌

Drop-in for Claude Code

Lives in your global CLAUDE.md. Claude reads the routing table and delegates automatically — no extra steps.

Free Providers

Provider	Model	Best for	Speed
Cerebras FREE	Llama 3.1 8B	Classification, scoring, yes/no	~284ms
Groq FREE	Llama 3.1 8B	Factual Q&A, general	~366ms
Gemini FREE	2.5 Flash Lite	Summarisation, translation, 1M context	~541ms
Mistral FREE	Mistral Small	Coding, creative writing	~1172ms
SambaNova FREE	Llama 3.3 70B	Deep reasoning, analysis	~1.4s
HuggingFace FREE	Llama 3 8B	Open-source fallback	~924ms
Pollinations NO KEY	Flux	Image generation	varies

Get Started

Clone the repo

git clone https://github.com/SoylentAquamarine/the-brain.git
cd the-brain && pip install -r requirements.txt

Add your API keys

Copy config/keys.example.json to config/keys.json and add free keys from Cerebras, Groq, Gemini, Mistral, and SambaNova.

Add to your global CLAUDE.md

python delegate.py --provider cerebras --type classification --prompt "Is this a question?"

Claude reads your CLAUDE.md routing table and delegates automatically.