Documentation
A condensed, link-out-rich overview. Full source-of-truth is the
README in liberzon/agents-gateway and the
interactive /docs page on your running instance.
Architecture
┌─────────────────────────────────────────────────────────────────┐
│ Client (web, CLI, IDE) │
└─────────────────────────────────┬───────────────────────────────┘
│ HTTPS / X-API-Key
▼
┌─────────────────────────────────────────────────────────────────┐
│ Agents Gateway (FastAPI) │
│ │
│ ┌─────────────┐ ┌────────────┐ ┌──────────────────────────┐ │
│ │ /v2/agents │ │ /v2/teams │ │ /v2/{prompts,knowledge,…}│ │
│ └──────┬──────┘ └─────┬──────┘ └──────────────┬───────────┘ │
│ └─────────┬─────┘ │ │
│ ▼ ▼ │
│ ┌─────────────────┐ ┌──────────────────────┐ │
│ │ Agno runtime │ │ Supervisor / Workers │ │
│ │ (LLM, tools, │ │ (Docker, K8s) │ │
│ │ sessions) │ └──────────────────────┘ │
│ └────────┬────────┘ │
└──────────────────┼──────────────────────────────────────────────┘
│
┌───────────┼─────────────┐
▼ ▼ ▼
PostgreSQL Qdrant LLM provider
(agents, (knowledge (OpenAI, Gemini,
sessions, vectors) Anthropic, …)
tokens,
queue)
API reference
All endpoints are documented on a running instance at /docs (OpenAPI). Summary:
| Resource | Endpoint | Purpose |
|---|---|---|
| Agents | GET/POST /v2/agents |
List, create, update, delete agents |
| Agent chat | POST /v2/agents/{id}/chat |
Chat with an agent (stream or non-stream) |
| Teams | GET/POST /v2/teams |
List, create, update, delete teams |
| Team run | POST /v2/teams/{id}/runs |
Execute a team |
| Knowledge | GET/POST /v2/knowledge/{tenant_id} |
Manage per-tenant knowledge entries |
| Tokens | GET/POST /v2/users/{user_id}/tokens |
OAuth token storage with auto-refresh |
| Prompts | GET/POST /v2/prompts |
Versioned prompt templates |
| Skills | GET/POST /v2/skills |
Reusable skill definitions |
| Engines | GET/POST /v2/engines |
Supervisor execution engines |
| Targets | GET/POST /v2/targets |
Supervisor run targets |
| Approvals | GET/POST /v2/approvals |
Pending approval review |
| Admin | GET/POST /admin/* |
API keys, cache management |
Authentication
X-API-Key: agw_xxxxxfor/v2/*X-Admin-Secret: ...for/admin/*AUTH_DISABLED=truefor local development
Configuration
Core environment variables — full list in the README:
| Variable | Required | Purpose |
|---|---|---|
DB_HOST / DB_PORT / DB_USER / DB_PASS / DB_DATABASE |
Yes | Postgres |
ADMIN_SECRET |
Yes (prod) | Admin endpoint auth |
GOOGLE_API_KEY / OPENAI_API_KEY / ANTHROPIC_API_KEY |
At least one | LLM provider |
QDRANT_URL |
No | Vector DB for knowledge |
PROMPT_STORAGE_BACKEND |
No | postgres (default) or service |
OTEL_TRACING_BACKEND / OTEL_LOGGING_BACKEND |
No | console, otlp, sentry, logtail |
AUTH_DISABLED |
No | Dev only — bypass auth |
Toolkits
| Toolkit | Providers | Tools |
|---|---|---|
| CalendarToolkit | Google Calendar, Microsoft Calendar | schedule_meeting, list_events, cancel_meeting |
| EmailToolkit | Gmail, Outlook | send_email, search_emails, create_draft |
| ContactsToolkit | Google Contacts, Microsoft Contacts | create_contact, search_contacts, list_contacts |
| DriveToolkit | Google Drive, OneDrive | list_files, read_file, upload_file |
All toolkits ship with confirmation-based workflows (user reviews before execution) and OAuth token management with auto-refresh.
Detail: toolkits/README.md.
Supervisor platform
Multi-agent execution with classification, job queues, approval flows, and containerized workers. Workers can run as Docker containers or Kubernetes pods, fed by an in-Postgres job queue.
Detail in the README and supervisor/, remote_agent/.
Database schema
| Schema | Tables | Purpose |
|---|---|---|
public |
agent_info, team_info, team_agent, knowledge_entries, user_tokens, api_keys, token_usage |
Core data |
prompts |
prompts |
Prompt templates |
ai |
(auto-created by Agno) | Agent sessions & memories |
Migration: db/migrations/setup.sql. Auto-applied on docker compose up.
Observability
OpenTelemetry-based, pluggable. Set OTEL_TRACING_BACKEND and OTEL_LOGGING_BACKEND:
# Sentry
OTEL_TRACING_BACKEND=sentry
SENTRY_DSN=https://xxx@sentry.io/xxx
# Cloud-native OTLP
OTEL_TRACING_BACKEND=otlp
OTEL_LOGGING_BACKEND=otlp
OTEL_OTLP_ENDPOINT=http://collector:4317
Development
./scripts/dev_setup.sh # Create venv + install deps
source .venv/bin/activate
./scripts/run_validate.sh # Format, lint, type-check
pytest tests/v2/ # Tests
./scripts/generate_requirements.sh upgrade # Bump deps
Detailed reference
Per-endpoint and per-subsystem deep dives (synced at build time from
liberzon/agents-gateway-docs):
Getting started
- Overview — architecture and key concepts
- Quickstart — setup and first API call
- Authentication — API keys, admin auth, dev mode
- Configuration — environment variables
API reference
- Agents —
/v2/agentsCRUD and chat - Teams — coordinate + supervisor team modes
- Prompts — versioned prompt templates
- Knowledge — tenant + collection-level
- Tokens — OAuth token storage
- Skills — reusable skill definitions
- Engines — execution engine registry
- Targets — execution target registry
- Approvals — HITL approval flow
- Admin — cache, API keys, background tasks
Supervisor platform
- Architecture — classification, worker types
- Agent packs — YAML pack format, CLI
- Remote agent — Docker/K8s container execution
- Toolkits — Calendar, Email, Drive, ClaudeCode, ManagedAgents
Reference
- Models — Pydantic request/response models
- Database schema — all tables
- LLM models — supported OpenAI, Gemini, Claude variants
- Deployment — docker-compose, production setup
- Development — dev setup, validation, testing