Docs

Orchet Docs

Build agents Orchet can discover, review, connect, and invoke.

Orchet is the orchestration layer for consumer apps. Developers publish capability contracts through the Orchet Store; users install and connect those agents from chat when a task needs them.

Build the quickstartUnderstand the contract

Architecture

Orchet is the orchestration layer, not your database.

Your app owns provider APIs, secrets, tables, migrations, queues, and business logic. Orchet stores the capability contract: tool names, object-shaped JSON Schema inputs, auth requirements, confirmation policy, trust state, and the locator needed to call your approved agent safely from chat.

You own

  • Provider API integrations and credentials
  • Private database schema and migrations
  • Execution code, retries, queues, and vendor-specific rules
  • Your deployment, incidents, and release cadence

Orchet owns

  • Agent Contract Registry snapshots
  • User install and chat-first connect flow
  • Planner argument validation and confirmation gates
  • Runtime invocation, rendering, and review state

Developer platform

SDK/API, remote MCP, and A2A normalize into one runtime.

The Agent Contract Registry is how Orchet knows which tools exist, what arguments they accept, which user credentials are required, and whether a tool can execute immediately or needs confirmation.

SDK/API

Best default

Host a small HTTPS agent with /.well-known/agent.json, /openapi.json, /health, and /tools/<tool> routes.

Remote MCP

Existing MCP server

Submit an MCP URL and OAuth metadata. Orchet snapshots tools/list and calls tools/call after review.

A2A

Peer agent

Submit an Agent Card when your service already behaves like an autonomous specialist agent.

Agent Contract Registry

The contract is what Orchet can safely plan against.

Every published agent needs at least one resolved tool with a stable operation id, a human-readable description, and an object-shaped JSON Schema input. The planner validates generated arguments against that approved snapshot before a tool call leaves Orchet.

JSON

{
  "agent_id": "weather-demo",
  "protocol": "sdk_api",
  "tools": [
    {
      "name": "weather_get_forecast",
      "description": "Get a forecast for a city.",
      "risk": "read",
      "requires_confirmation": "none",
      "input_schema": {
        "type": "object",
        "required": ["city"],
        "properties": {
          "city": { "type": "string" },
          "units": { "type": "string", "enum": ["metric", "imperial"] }
        }
      }
    }
  ]
}

Quickstart path

Build, validate, deploy, submit, test from chat.

1

Create account

Sign in at /developer so Orchet can attach submissions to your developer identity.

2

Create developer key

Generate ORCHET_DEVELOPER_TOKEN at /developer/keys for CLI submissions.

3

Scaffold

npx create-orchet-agent init weather-demo

4

Deploy

Host manifest, OpenAPI, health, and tool routes on HTTPS.

5

Submit manifest

Use orchet-agent submit / submit-mcp / submit-a2a or the developer platform.

6

Test from chat

After review and approval, users connect from chat when a task needs your agent.

Submission contract

What reviewers check before approval.

Review looks for stable discovery endpoints, resolved tool schemas, minimal OAuth scopes, explicit confirmation policy for write or side-effect tools, cancellation semantics for obligations, and clear user-facing copy. Provider credentials never belong in the submission payload.

Submission guideReview policyPublic launch gatesTesting checklist