API Reference

This content is not available in your language yet.

This section is a complete, field-by-field reference for every query and mutation exposed by the Wetel API. It complements the conceptual guides elsewhere in these docs (start with Getting Started, Authentication, and Multi-Tenancy if you haven’t yet) by documenting exact argument lists, return shapes, and realistic request/response examples for each operation.

The API is GraphQL-only

There is a single endpoint for all operations:

https://api.wetel.dev/graphql

There is no separate REST API for the core platform (agents, sessions, workflows, tenants, and so on). A small number of avatar/media operations are exposed over REST by a companion service — see Avatar Service REST API for those.

This reference is kept in sync with the schema, but the schema is the ultimate source of truth. Point any GraphQL client’s introspection query at the live endpoint to see the exact current shape at any time:

{
  __schema {
    types {
      name
    }
  }
}

Most GraphQL clients (Apollo Sandbox, Insomnia, Postman, GraphiQL) also have a built-in docs panel that reads live introspection data — use it to browse arguments and return types interactively while you build.

Request limits

Every query/mutation is subject to two structural limits, enforced server-side regardless of authentication method:

Query depth: max 7. Counted as nested selection-set levels — e.g. session { messages { ... } } is depth 2. A query exceeding this is rejected before execution with a GraphQL validation error, not partially run.
Query complexity: max 50. A cost estimate based on field count and list-argument sizes (e.g. a first: 100 connection argument costs more than first: 10). Exceeding this is also a pre-execution rejection.

Neither limit affects introspection itself or any realistic query this API’s own resolvers produce — the schema has minimal nesting today (one Relay connection type, no deep object graphs), so a client following normal usage patterns won’t hit either ceiling. They exist to bound worst-case resource cost from a maliciously or accidentally deep/expensive request, not to constrain legitimate integration queries.

Reference index

Authentication — login, registration, password reset, two-factor auth, current user profile
Tenant — reading and updating your own tenant, API key generation
Agents — creating and configuring AI agent personas
Sessions — starting conversations, sending messages, live events
Workflows — the visual conversation graph engine
Channels — connecting an agent directly to Telegram (WhatsApp/Slack/Lark planned)
MCP & Custom Actions — connecting third-party tools to your agents
Knowledge Base & Embeddings — RAG document ingestion and retrieval
LLM API — call an LLM directly, single-response or streaming, no session/agent/workflow needed
Evaluation & Export — post-session scoring and document export
Avatar Config — tenant-level avatar/voice defaults
Admin Resources — roles, permissions, notifications, media, banners
Avatar Service REST API — the companion REST surface for avatar/TTS/STT operations