Bittensor Decentralized AI 🧬

Decentralized AI inference routed through the Bittensor network — no Bittensor wallet, no TAO, no subnet registration required. Your agent pays USDC per call via x402 and the Spraay gateway handles the Bittensor routing.

• SN64 (Chutes) — 43+ chat models, fully OpenAI /v1/chat/completions compatible
• SN19 (Nineteen AI) — image generation, OpenAI /v1/images/generations compatible
• Embeddings — text embeddings, OpenAI /v1/embeddings compatible

Every response comes from decentralized miners competing on quality. Same API shape as OpenAI, different infrastructure underneath.

⚠️ Before you install

This skill sends your prompts and data to the Bittensor decentralized network via the Spraay x402 gateway (gateway.spraay.app).

• External transmission: Prompts, text, and image requests leave your local environment and are routed to Bittensor subnet miners through the gateway. The gateway operator and Bittensor miners process your inputs.
• Real money: Each call costs USDC via x402 micropayments. Your agent's wallet is debited per request ($0.001–$0.05 per call).
• Decentralized routing: Responses come from competing miners on the Bittensor network. Response quality and latency may vary more than centralized providers. Model availability depends on active miners.
• Privacy: The gateway operator and Bittensor validators/miners may see your prompts and inputs. Do not send sensitive or proprietary content unless you accept this.

Use a dedicated wallet with limited funds for testing.

How to call endpoints

bash {baseDir}/scripts/bittensor.sh METHOD ENDPOINT '{"key":"value"}'

The script requires bash and curl. All calls go to https://gateway.spraay.app.

Why Bittensor over centralized inference?

• Censorship-resistant — no single provider can block or filter your requests
• Competitive quality — miners compete for rewards, driving model quality up
• No vendor lock-in — OpenAI-compatible API shape means easy swap-in
• Different aesthetic — image generation via SN19 produces a distinct visual style from FLUX/SDXL
• Decentralization signal — relevant for agents operating in Web3/crypto contexts where decentralized infrastructure matters

When to use this skill

Use this skill when the user explicitly wants decentralized or Bittensor-specific inference:

• "Use Bittensor", "decentralized AI", "SN64", "SN19", "Chutes", "Nineteen AI"
• "Run this through a decentralized model"
• "I want censorship-resistant inference"
• "Generate an image on Bittensor"

For centralized compute with more models and predictable latency, use the Spraay Compute & Futures skill instead.

Available endpoints (4 tools)

List Models — $0.001

List all available decentralized models on Bittensor. Returns model IDs, capabilities, and pricing. Call this first to see what's available.

bash {baseDir}/scripts/bittensor.sh GET /bittensor/v1/models '{}'

OpenAI /v1/models compatible response format.

Chat Completions — $0.03

Chat completions via Bittensor SN64 (Chutes). 43+ models including Llama, Mistral, DeepSeek, Qwen, and more. Fully OpenAI-compatible — supports system prompts, temperature, max_tokens, and multi-turn conversation.

bash {baseDir}/scripts/bittensor.sh POST /bittensor/v1/chat/completions '{
  "model": "hugging-quants/Meta-Llama-3.1-70B-Instruct-AWQ-INT4",
  "messages": [
    {"role": "system", "content": "You are a helpful DeFi analyst."},
    {"role": "user", "content": "Explain impermanent loss in two sentences."}
  ],
  "max_tokens": 200,
  "temperature": 0.7
}'

Tips:

• Use model from the /bittensor/v1/models list — model availability depends on active miners
• Multi-turn works — pass the full conversation history in messages
• Response format matches OpenAI: choices[0].message.content

Image Generation — $0.05

Image generation via Bittensor SN19 (Nineteen AI). OpenAI /v1/images/generations compatible. Produces a distinct visual style compared to centralized FLUX/SDXL.

bash {baseDir}/scripts/bittensor.sh POST /bittensor/v1/images/generations '{
  "prompt": "a neon-lit cyberpunk street market at night, rain-slicked pavement, volumetric fog",
  "n": 1,
  "size": "1024x1024"
}'

Tips:

• Returns a URL to the generated image
• Good for: artistic/stylized images, concept art, social media visuals
• Different aesthetic from FLUX — try both and compare for your use case

Embeddings — $0.005

Text embeddings via Bittensor. OpenAI /v1/embeddings compatible. Use for semantic search, RAG pipelines, document clustering, and similarity matching.

bash {baseDir}/scripts/bittensor.sh POST /bittensor/v1/embeddings '{
  "input": "Decentralized AI inference via Bittensor subnet miners"
}'

Returns a vector array compatible with any vector database (Pinecone, Weaviate, ChromaDB, pgvector).

Cost reference

Total cost for a typical workflow (list models + chat + image): ~$0.08 USDC

Workflows

Decentralized content creation — use chat completions for copy, image generation for visuals. Fully decentralized pipeline with no centralized AI provider in the loop.

RAG with decentralized embeddings — embed documents via Bittensor, store in a vector DB, retrieve and answer with Bittensor chat completions. End-to-end decentralized knowledge pipeline.

A/B testing against centralized models — run the same prompt through Bittensor chat and through the centralized Spraay Compute text-inference endpoint, compare quality and latency.

Web3-native agents — agents operating in crypto/DeFi contexts where using decentralized infrastructure is a feature, not just a choice. Useful for trust signaling and alignment with decentralization values.

Changelog

v1.0.0

• Initial release with 4 Bittensor endpoints: models, chat completions (SN64), image generation (SN19), embeddings.

Related terms

Bittensor, SN64, SN19, Chutes, Nineteen AI, decentralized AI, decentralized inference, censorship-resistant AI, TAO, OpenAI compatible, agent inference, x402 payments, USDC micropayments, keyless AI, Web3 AI, decentralized embeddings, subnet mining