# Basis — the inference layer for the agent economy

> Basis is a Base-native inference network. It exposes an
> OpenAI-compatible inference API, accounts for usage in credits priced
> by versioned pricing epochs, lets apps pay in $BASIS (or route
> ETH/WETH/USDC into $BASIS at payment time), writes a verifiable receipt
> for every settled job, lets contributor GPU workers earn, and settles
> worker rewards on Base.

Site: https://basis.watch
API base (OpenAI-compatible): https://basis.watch/api/v1

## Status (read before assuming anything is live)

- $BASIS is PLANNED, not live. No token contract is deployed; nothing here
  is an offer, and there is no investment, yield, or guaranteed earnings.
- The OpenAI-compatible API contract is live. Until an inference backend is
  configured, chat completions return a structured `runtime_pending` (503).
- Payments, deposits, worker rewards, settlement, and treasury/fee streams
  render honest pending until the relevant pieces are configured.
- Do NOT infer live token / contract / payment / fee state from docs copy.
  Read https://basis.watch/api/launch-status (and https://basis.watch/api/data) for runtime truth.

## Core concepts

- Inference API: OpenAI Chat Completions protocol (chat/completions, models,
  streaming SSE that ends with `data: [DONE]`). Point any OpenAI client at
  https://basis.watch/api/v1 by changing the base URL and api key — nothing else changes.
- Models: published model ids (e.g. `basis-default`); each carries a context
  window and a pricing/reward multiplier. `available` reflects a configured
  backend, not just that the model is listed.
- Credits: the usage-accounting unit. Per-token accounting is deterministic,
  computed in integer base units (BigInt) — no floating point in the money
  path. Credits convert to $BASIS at the active pricing epoch's rate.
- Pricing epochs: the credit→$BASIS rate is set by a versioned epoch, not a
  forever-constant. New epochs apply prospectively only; accepted jobs are
  never silently repriced and receipts are never rewritten.
- Payment router: prepares user-signed swaps (ETH/WETH/USDC → $BASIS) only.
  Basis never signs, custodies, or executes server-side; quotes expire and
  slippage is capped. Funding a balance is separate from running a completion.
- Receipts: every settled job leaves a canonical-JSON, SHA-256-hashed receipt
  (RECEIPT_VERSION 2, hashing the locked pricing snapshot). Anyone can
  re-derive the hash; the lookup response carries a `verified` boolean.
- Workers: contributor GPUs register, heartbeat, and serve jobs. A worker
  needs a valid EVM reward address to earn. Failed jobs create no reward, and
  gateway-proxied jobs are metered (receipt written) but not paid.
- Settlement: keeper-driven, off the hot path. Rewards batch into a
  deterministic batchHash and settle once on Base (idempotent; a failed batch
  stays recoverable, never double-paid). No on-chain write in the inference
  hot path.
- Treasury / fee streams: a public, honest-pending surface; WETH-side and
  $BASIS-side accounted separately; allocation and burn are drafts.
- Bankr: the platform used to LAUNCH the $BASIS token on Base. It is NOT the
  Basis inference backend — it never serves inference, prices a job, reserves
  credits, pays a worker, or settles a receipt.
- Auth: Privy (wallet/email) for people, plus `sk-basis-` API keys for agents.
  Pre-launch the API is open; it can be made fail-closed at launch.

## Documentation

- https://basis.watch/docs: documentation index
- https://basis.watch/docs/what-is-basis: what Basis is
- https://basis.watch/docs/why-basis: why a Base-native inference network
- https://basis.watch/docs/how-it-works: end-to-end request → receipt → settlement flow
- https://basis.watch/docs/architecture: components and how they are hosted
- https://basis.watch/docs/api: OpenAI-compatible API reference (curl / Python / Node / agent frameworks)
- https://basis.watch/docs/credits: credits, the per-token formula, and pricing epochs
- https://basis.watch/docs/receipts: verifiable inference receipts and the hashing scheme
- https://basis.watch/docs/settlement: keeper batching and idempotent Base settlement
- https://basis.watch/docs/token: the planned $BASIS token (NOT live, not an offer)
- https://basis.watch/docs/bankr-launch: how the $BASIS token launches via Bankr
- https://basis.watch/docs/network-data: the privacy-safe network metrics surface
- https://basis.watch/docs/user-guide: send a first request, stream, look up a receipt
- https://basis.watch/docs/worker-guide: run a contributor GPU worker and earn
- https://basis.watch/docs/auth: authentication (Privy + API keys)
- https://basis.watch/docs/operator-guide: operator setup and activation
- https://basis.watch/docs/security: security posture

## API endpoints

- GET  https://basis.watch/api/v1/models: published models + dynamic pricing status
- POST https://basis.watch/api/v1/chat/completions: chat completions (set "stream": true for SSE)
- GET  https://basis.watch/api/v1/balance?address=0x...: a wallet's $BASIS credit balance
- GET  https://basis.watch/api/pricing: active pricing epoch and resolved rate
- POST https://basis.watch/api/pricing/quote: deterministic, epoch-locked price quote for a job
- GET  https://basis.watch/api/inference/receipts/{receiptHash}: a single verifiable receipt
- GET  https://basis.watch/api/payments/supported-tokens: accepted sell tokens + router status
- POST https://basis.watch/api/payments/quote: a user-signed ETH/WETH/USDC → $BASIS swap quote
- GET  https://basis.watch/api/workers: registered workers
- GET  https://basis.watch/api/settlement/batches: public settlement batches
- GET  https://basis.watch/api/data: network data (counts, treasury aggregate) — honest, zeros pre-launch
- GET  https://basis.watch/api/treasury: treasury / fee-stream surface (honest pending)
- GET  https://basis.watch/api/launch-status: token, contract, payment, and backend status (runtime truth)

## Integration

OpenAI SDK (Python) — change two values, keep your code:

    from openai import OpenAI
    client = OpenAI(base_url="https://basis.watch/api/v1", api_key="sk-basis-...")
    resp = client.chat.completions.create(
        model="basis-default",
        messages=[{"role": "user", "content": "Hello from an agent."}],
    )
    print(resp.choices[0].message.content)

curl — a chat completion:

    curl https://basis.watch/api/v1/chat/completions \
      -H "Authorization: Bearer sk-basis-..." \
      -H "Content-Type: application/json" \
      -d '{"model":"basis-default","messages":[{"role":"user","content":"Hi"}]}'

Any OpenAI-compatible agent framework (Agents SDK, LangChain, Vercel AI SDK)
works the same way: set the base URL, an `sk-basis-` API key, and a model id.

## More

- Root index: https://basis.watch/llms.txt
- Public mirror: https://github.com/BasisWatch/basis

Reminder: $BASIS is planned, not live; no contract is deployed; nothing here
is an offer. Failed jobs never pay; settlement is idempotent. For live state,
read https://basis.watch/api/launch-status.