Integration overview
Three concrete paths for using Relay-E in your product today.
If you want to add an AI feature to your product, here are the three ways people actually do it. Pick by the column you fit into.
Path 1 — Pull the Docker image
Best for: SaaS apps, internal tools, anything where you want to "just have an AI assistant" without maintaining the engine yourself.
services:
relay-e:
image: ghcr.io/belulok/relay-e:latest
ports: ["3001:3001"]
environment:
DEV_API_KEY: ${RELAY_E_KEY}
DATABASE_URL: postgres://postgres:postgres@postgres:5432/relay_e
REDIS_URL: redis://redis:6379
OPENROUTER_API_KEY: ${OPENROUTER_API_KEY}
depends_on: [postgres, redis]
relay-e-worker:
image: ghcr.io/belulok/relay-e:latest
command: ["npm", "run", "worker"]
environment:
REDIS_URL: redis://redis:6379
DATABASE_URL: postgres://postgres:postgres@postgres:5432/relay_e
depends_on: [redis]Then call the HTTP API. Pros: no fork, ~5-min setup. Cons: only the bundled financial-advisor skill until skill-as-config lands; for real domain logic, see Path 2.
Path 2 — Fork and customize
Best for: production deployments with real domain logic — your own tools, skills, and connectors.
git clone https://github.com/belulok/relay-e.git my-relay-e
cd my-relay-e
# Edit apps/api/src/bootstrap/registries.ts — add your skills and tools
# Connect to your data (Guides → Connecting your data)
git push origin master --tags
# Your fork's release.yml builds ghcr.io/yourorg/my-relay-e:vX.Y.ZYou inherit the agent loop, OpenAPI surface, queue, multi-tenancy, and observability — and add your domain.
Path 3 — Just the HTTP API (any language)
Best for: integrating from a language other than TypeScript. Works the same as Stripe / OpenAI APIs.
# Generate a typed client from the OpenAPI spec
npx openapi-typescript-codegen --input http://localhost:3001/openapi.json --output ./client # TS
openapi-python-client generate --url http://localhost:3001/openapi.json # Python
oapi-codegen -package relayE http://localhost:3001/openapi.json > relay_e.go # GoEvery language has an OpenAPI generator. The spec is auto-updated as you add routes.
Decision tree
Are you in a hurry / want a demo?
├─ Yes → Path 1 (Docker, 5 min)
└─ No
Do you need your own skills/tools/connectors?
├─ No → Path 1
└─ Yes
Are you on TypeScript?
├─ Yes → Path 2 (fork)
└─ No → Path 2 (fork) for the engine + Path 3 (codegen) for your clientWhat's missing today
| Need | Status | Workaround |
|---|---|---|
| First-class TS/Python SDK | Not yet — packages/sdk-ts is roadmap | OpenAPI codegen (above) |
| Skill-as-config (no fork required) | Roadmap | Fork + edit bootstrap/registries.ts |
| Production auth (per-tenant keys, OAuth) | Dev-only DEV_API_KEY | Front it with your auth proxy |
Hosted version (relay-e.com) | Not yet | Self-host via Docker on Fly / Railway / your VPC |
| Real built-in connectors (Stripe, Plaid) | Stub MCP only | Roll your own MCP server or REST adapter |
These are roadmap items. Once the SDK + skill-as-config + hosted tier land, the integration story tightens to "npm install @relay-e/sdk + paste your API key."