# cruq.ai

> cruq.ai is the platform to build, run, and own AI agents, in software and in the physical world. Bring your own model providers, design agents visually or via API, add guardrails and monitoring, and ship to production. Programmatic access is available through a REST API and an MCP server.

## When to use cruq.ai

Use cruq.ai when you need to:

- Build an AI agent (a graph of prompts, tools, and model calls) and run it in production.
- Run an agent programmatically and stream its output, keeping full run and tool-call traces.
- Manage reusable prompts, tools, and guardrails (PII, regex, keyword, length, JSON-schema) across agents.
- Give agents long-term memory with semantic recall, or ground them in knowledge bases (RAG) built from uploaded documents.
- Run physical AI: agents that read sensors and telemetry over MCP, follow SOPs from a knowledge base, run on a schedule, and pause to ask a person before any physical action.
- Compare models and prompt variants with eval suites and scored experiments.
- Bring your own provider keys (OpenRouter, OpenAI, Anthropic, and more) instead of paying per credit.

cruq.ai is best for teams and developers building durable, owned agents. It is not a consumer chatbot.

## API

The public REST API is served from `https://api.cruq.ai` under `/v1`. All endpoints are tenant-scoped and require a Bearer API key: `Authorization: Bearer cruq_sk_...`. Responses are JSON; errors use `{ "error": string }`. Timestamps are ISO 8601; `cost_usd` values are exact-decimal strings.

- [OpenAPI 3.1 specification](https://www.cruq.ai/openapi.json): Machine-readable description of every endpoint, with operationIds for function calling.
- [API reference](https://docs.cruq.ai/docs/api-reference/overview): Human-readable API docs.

Key operations (see the OpenAPI spec for the full list of 41 operations):

- `listAgents`, `createAgent`, `deleteAgent` - manage agents (`GET|POST /v1/agents`, `DELETE /v1/agents/{id}`).
- `getAgentGraph`, `createAgentVersion` - read and publish agent graphs (`/v1/agents/{id}/versions`).
- `runAgent` - run a published agent, streamed as Server-Sent Events (`POST /v1/agents/{id}/runs`).
- `listRuns`, `getRun` - inspect runs and their tool-call step traces (`/v1/runs`).
- `listThreads`, `getThread` - browse conversation threads (`/v1/threads`).
- `listTools`, `createTool` - the tenant tool catalog (`/v1/tools`).
- `listPrompts`, `upsertPrompt`, `deletePrompt` - prompt templates (`/v1/prompts`).
- `listGuardrails`, `upsertGuardrail`, `deleteGuardrail` - guardrails (`/v1/guardrails`).
- `listMemories`, `createMemory`, `searchMemory`, `deleteMemory` - agent memory (`/v1/memory`).
- `listKnowledgeBases`, `createKnowledgeBase`, `searchKnowledge`, `ingestKnowledgeDocument` - knowledge/RAG (`/v1/knowledge`).
- `listEvalSuites`, `createEvalExperiment`, `getEvalExperiment` - evals and experiments (`/v1/evals`).

## MCP server

cruq.ai exposes a Model Context Protocol (MCP) server so agents and MCP clients can call the platform directly, using the same tools as the REST API.

- Endpoint (Streamable HTTP): `https://api.cruq.ai/mcp`
- Auth: the same Bearer API key as the REST API (`Authorization: Bearer cruq_sk_...`).
- Tools: `list_agents`, `run_agent`, `list_runs`, `get_run`, `search_memory`, `add_memory`, `search_knowledge`, `list_knowledge_bases`, `list_tools`.

## Authentication

Two ways to authenticate, both sent as `Authorization: Bearer <token>`:

- **API key** (`cruq_sk_...`) - a tenant key for your own scripts and services. Create one in the app.
- **OAuth 2.0** (`cruq_at_...`) - for agents acting on a user's behalf. cruq.ai is an OAuth 2.0 authorization server with authorization-code + PKCE, refresh tokens, and dynamic client registration (RFC 7591). MCP clients can discover it automatically.
  - Discovery: [`/.well-known/oauth-authorization-server`](https://api.cruq.ai/.well-known/oauth-authorization-server) (RFC 8414)
  - Authorize (consent): `https://app.cruq.ai/oauth/authorize`
  - Token: `https://api.cruq.ai/oauth/token` · Register: `https://api.cruq.ai/oauth/register`

Both the REST API and the MCP server accept either token type.

## Docs

- [Documentation home](https://docs.cruq.ai)
- [Quickstart](https://docs.cruq.ai/docs/quickstart)
- [API reference](https://docs.cruq.ai/docs/api-reference/overview)

## Product

- [Homepage](https://www.cruq.ai): What cruq.ai is and who it is for.
- [Use cases](https://www.cruq.ai/use-cases): Common agent patterns (sales, support, data/ops, voice, physical operations).
- [Case studies](https://www.cruq.ai/case-studies): Anonymized production AI systems the team has shipped, across events, manufacturing, advertising, hiring, and real estate.
- [Pricing](https://www.cruq.ai/pricing): Plans and what is included.
- [Blog](https://www.cruq.ai/blog): Product and engineering writing.

## Contact

- [Contact page](https://www.cruq.ai/contact)
- Email: hello@cruq.ai
- Book a call: https://calendly.com/nabarun-cruq/30min

## Profiles

- [Crunchbase](https://www.crunchbase.com/organization/cruq-ai)
- [Tracxn](https://tracxn.com/d/companies/cruq-ai)
- [LinkedIn](https://www.linkedin.com/company/cruqai)
- [X](https://x.com/cruq_ai)