# CrossCompute

> CrossCompute is an AI inference gateway for both human teams and autonomous agents. It provides OpenAI-compatible inference, API key quota controls, and x402 stablecoin credit top-ups.

## Human Console

- Dashboard: https://www.crosscompute.io/app.html
- Human flow: Google sign-in -> bootstrap account -> manage `cc_live_*` keys -> call `https://www.crosscompute.io/v1`

## Agent Runtime

- Inference API base: https://www.crosscompute.io/v1
- Agent key type: `cc_agent_*`
- Agent flow: register -> usage -> top-up -> inference
- Safe default: non-spending unless `CROSSCOMPUTE_ALLOW_SPEND=true`
- Payment boundary: only `POST /api/agents/topup` is charging

## Core Agent Endpoints

- `POST /api/agents/register`: Create or update an agent account and issue a `cc_agent_*` API key.
- `GET /api/agents/me`: Return current agent account and key summary.
- `GET /api/agents/usage`: Return purchased, used, and remaining credits.
- `POST /api/agents/topup`: Top up agent credits through x402 payment flow.
- `GET /v1/models`: Local OpenAI-style model catalog generated by CrossCompute.
- `POST /v1/chat/completions`: OpenAI-compatible inference.

OpenAI compatibility notes:

- `role="developer"` is accepted and normalized to `system` where needed.
- Content parts `input_text` and `input_image` are accepted.
- `max_completion_tokens` is accepted and normalized for non-OpenAI upstream providers.

## Core Human API Endpoints

- `POST /api/bootstrap-account`: Create user profile and initial API key.
- `GET /api/keys`: List user keys.
- `POST /api/keys`: Create user keys.
- `POST /api/keys/:id/deactivate`: Deactivate a key.

## Model Pool (Routing-Behind)

- DeepSeek series: `deepseek-chat`, `deepseek-reasoner`
- Kimi series: `moonshot/kimi-k2`, `moonshot/kimi-k2.5`, `moonshot/kimi-k2-turbo-preview`
- Qwen series: `qwen/qwen-max`, `qwen/qwen-plus`, `qwen/qwen-flash`, `qwen/qwen3-max`
- Gemini series: `google/gemini-2.5-pro`, `google/gemini-2.5-flash`, `google/gemini-2.5-flash-lite`
- Claude series: `anthropic/claude-sonnet-4.6`, `anthropic/claude-opus-4.6`, `anthropic/claude-haiku-4.5`
- Grok series: `xai/grok-4`, `xai/grok-4-fast-reasoning`, `xai/grok-code-fast-1`
- OpenAI GPT series: `openai/gpt-5.4`, `openai/gpt-5-mini`, `openai/o3-mini`
- MiniMax series: `minimax/m2.5`, `minimax/m2.5-highspeed`, `minimax/m2.1`
- Llama series: `llama/llama-4-maverick`, `llama/llama-3.3-70b`, `llama/llama-3.1-8b`

## Routing Profiles

- Model alias accepts `auto`, `cheap`, `medium`, `premium` (legacy `eco` is also accepted)
- `auto` performs adaptive model ranking by quality/cost/latency/runtime-health

## Routing Runtime Defaults (Current)

- `PROXY_ROUTING_DEFAULT_PROFILE=auto`
- `PROXY_ROUTING_MAX_FALLBACKS=5`
- `PROXY_ROUTING_ENABLE_RUNTIME_HEALTH=true`
- `PROXY_ROUTING_CIRCUIT_BREAKER_FAILURES=3`
- `PROXY_ROUTING_CIRCUIT_BREAKER_COOLDOWN_MS=120000`
- `PROXY_ROUTING_HEALTH_ALPHA=0.2`
- `PROXY_ROUTING_CHEAP_MAX_USD=0` (disabled)
- `PROXY_ROUTING_MEDIUM_MAX_USD=0` (disabled)
- `PROXY_ROUTING_PREMIUM_MAX_USD=0` (disabled)
- `PROXY_ROUTING_EXPLORATION_EPSILON=0` (disabled)

## Agent Model Selection Rules

- If your goal is lowest cost, use `model="cheap"`.
- If your goal is balanced quality/cost, use `model="medium"`.
- If your goal is best quality, use `model="premium"`.
- If you need deterministic provider/model, use explicit IDs like:
  - `openai/gpt-5-mini`
  - `anthropic/claude-sonnet-4-6`
  - `deepseek-chat`

Anthropic naming note:

- Prefer `claude-sonnet-4-6` (hyphen) rather than `claude-sonnet-4.6` (dot).

## Response Headers For Agent Introspection

- `x-crosscompute-routing-profile`
- `x-crosscompute-routing-tier`
- `x-crosscompute-routing-budget`
- `x-crosscompute-routing-version`
- `x-crosscompute-routing-intelligence`
- `x-crosscompute-routing-intelligence-session`
- `x-crosscompute-routing-intelligence-request-id`
- `x-crosscompute-routed-model`
- `x-crosscompute-provider`
- `x-crosscompute-price-rule`
- `x-crosscompute-billed-credits`

Important:

- `x-crosscompute-routing-profile` is a response header for introspection, not a request parameter.
- To choose cheap routing, set `model=\"cheap\"` in `POST /v1/chat/completions` JSON body.

## Machine-Readable Docs

- Agent quickstart: https://www.crosscompute.io/agent-skill.md
- Full API parameter registry: https://www.crosscompute.io/api-parameters.md
- Crawler access policy: https://www.crosscompute.io/crawler-access-policy.md
- Agent executable quickstart script: https://www.crosscompute.io/examples/crosscompute-agent-quickstart.mjs
- Agent trust policy: https://www.crosscompute.io/.well-known/agent-trust-policy.json
- Security contact: https://www.crosscompute.io/.well-known/security.txt
- OpenAPI spec: https://www.crosscompute.io/openapi.json
- Plugin manifest: https://www.crosscompute.io/.well-known/ai-plugin.json
- Robots: https://www.crosscompute.io/robots.txt
- Sitemap: https://www.crosscompute.io/sitemap.xml

## Payment Notes

- x402 top-up supports challenge-response flow.
- Production target is Base mainnet USDC (`eip155:8453`).
- Testnet override is supported with `X402_REQUIRE_MAINNET=false` + `X402_NETWORK=eip155:84532`.
- Credits from top-up are added to the calling API key quota.

## Crawler Notes

- Crawl/index is intended for docs and machine-readable policy pages.
- Authenticated surfaces (`/api/*`, `/v1/*`, `/app.html`) are non-index targets.
- Cloudflare Browser Rendering `/crawl` compatibility is documented at:
  - https://www.crosscompute.io/crawler-access-policy.md