Gateway

The v1 gateway routes incoming requests to registered plugins. It supports three payment methods — Stripe (existing), MPP (Machine Payments Protocol) via the Tempo blockchain, and session-based billing for low-latency, off-chain per-call payments.

Base URL

POST /api/v1/gateway

Request

Headers

Header	Type	Required	Description
`Content-Type`	string	Yes	Must be `application/json`
`X-Plugin-Id`	string	No	Plugin to route the request to. Overrides the `plugin` field in the body.
`X-Payment-Method`	string	No	Payment method to use: `stripe`, `mpp`, or `session`. Defaults to `stripe`.
`Authorization`	string	No	For MPP payments, use `Payment <credential>`. For session auth, handled via cookies.
`X-Session-Id`	string	No	Active session ID for session-based billing. Required when `X-Payment-Method` is `session`.
`X-Wallet-Address`	string	No	Wallet address that owns the session (0x-prefixed). Required when `X-Payment-Method` is `session`.
`Accept`	string	No	Set to `text/event-stream` for streaming responses (where supported).

Body

{
  "plugin": "agent",
  "messages": [{ "role": "user", "content": "Hello" }]
}

Field	Type	Required	Description
`plugin`	string	No	Plugin ID to route to. The `X-Plugin-Id` header takes priority if both are provided. Defaults to `agent`.

Additional body fields are forwarded to the target plugin.

Plugins

The gateway routes to the following plugins:

Plugin ID	Name	Description	Auth required	Streaming
`agent`	Agent	Agent orchestrator for multi-step tasks	No	Yes
`generate-text`	Text Generation	LLM text generation	No	Yes
`tts`	Text-to-Speech	Speech synthesis	No	No
`stt`	Speech-to-Text	Speech transcription	No	No

agent is the default plugin when no plugin ID is specified.

Authentication

No plugins currently require authentication. The gateway checks whether a plugin has its auth flag enabled — since no built-in plugin sets this flag, all requests are processed without requiring a session, MPP credential, or payment session. If a custom plugin is registered with auth: true, the gateway enforces either a valid session (cookie-based via NextAuth), a verified MPP payment credential, or an active payment session. If you pay with MPP or use session-based billing, cookie-based authentication is not required for auth-enabled plugins.

Payment flow

The gateway supports three payment methods per request:

Stripe — Default. Requires an active subscription or credits. See Stripe integration.
MPP — Crypto-native payments on the Tempo blockchain. See MPP payments.
Session — Off-chain, per-call billing using a pre-funded payment session. See MPP payments — sessions and the wallet sessions API.

The server selects the payment method using the following priority:

X-Payment-Method header (session, mpp, or stripe)
Presence of an Authorization: Payment header (implies mpp)
Default: stripe

MPP 402 challenge

When an MPP request has no valid credential, the gateway returns 402 Payment Required with pricing information for both payment methods:

{
  "error": "payment_required",
  "message": "Payment required for agent. Choose payment method: Stripe or Tempo MPP.",
  "mpp": {
    "scheme": "Payment",
    "amount": "0.05",
    "currency": "0x20c0000000000000000000000000000000000000",
    "recipient": "0xd8fd0e1dce89beaab924ac68098ddb17613db56f",
    "description": "Agent orchestrator request",
    "nonce": "a1b2c3d4e5f6...",
    "expiresAt": 1742472000000
  },
  "stripe": {
    "checkoutUrl": "/api/v1/payments/stripe/create?plugin=agent",
    "amount": "0.05",
    "currency": "usd"
  }
}

The WWW-Authenticate header is also set:

Payment amount="0.05", currency="0x20c0000000000000000000000000000000000000", recipient="0xd8fd0e1dce89beaab924ac68098ddb17613db56f"

Session-based billing

When X-Payment-Method is session, the gateway auto-debits the caller’s payment session using an off-chain voucher. This avoids the 402 challenge/response round-trip and settles each call in sub-100ms. To use session-based billing:

Open a payment session via POST /api/wallet/sessions. See wallet sessions.
Include X-Session-Id and X-Wallet-Address headers on every gateway request.
The gateway looks up the session, verifies the balance covers the plugin price, and debits the session automatically.
The response includes a Payment-Receipt header with the voucher reference and an X-Session-Remaining header with the updated balance.

If the session is missing, expired, or has insufficient balance, the gateway returns 402 with a descriptive error. See error responses for the full list of session-related error codes.

Response

Success

The gateway forwards the request body to the matched plugin’s upstream URL and returns the plugin’s response directly. The HTTP status code and Content-Type header are preserved from the upstream response.

The gateway no longer returns a synthetic response with plugin, message, timestamp, and payment fields. Responses are now proxied directly from the upstream plugin service. The response shape depends entirely on the plugin being called. Requests have a 30-second timeout.

When paid via MPP, the response includes a Payment-Receipt header with the transaction hash. When paid via a session, the Payment-Receipt header contains the voucher reference (formatted as session:<sessionId>:<nonce>).

Response headers

Header	Description
`x-plugin-id`	The plugin that handled the request
`Payment-Receipt`	Payment receipt. For MPP: transaction hash. For sessions: voucher reference (`session:<sessionId>:<nonce>`). Only present for MPP or session-paid requests.
`X-Session-Remaining`	Remaining session balance in USD after the debit. Only present for session-paid requests.

Error responses

Status	Error code	Description
400	`unknown_plugin`	No pricing configured for the requested plugin (session billing only)
401	`Unauthorized`	Authentication required for a protected plugin and no valid session, MPP credential, or payment session
402	`payment_required`	MPP payment required. Response includes challenge and pricing.
402	`session_required`	Session billing was requested but `X-Session-Id` or `X-Wallet-Address` header is missing
402	`session_invalid`	No active session found for the provided session ID and wallet address
402	`insufficient_balance`	Session balance is too low to cover the plugin price. Response includes `session.remaining` and `session.cost`.
402	`voucher_failed`	The off-chain voucher could not be processed (e.g., concurrent debit race)
500	`internal`	Internal server error
502	`no_plugin`	No plugin registered for the given ID

Gateway chat proxy

POST /api/gateway/chat

Sends a chat message to the caller’s deployed agent by enqueuing it as a job on the backend control plane. The server validates the user’s agent, applies rate limiting and workload gating, and returns a job ID that you poll for the result.

This endpoint no longer connects to the gateway via WebSocket. Chat messages are enqueued as asynchronous jobs on the backend. Use the returned jobId to poll for the result via GET /api/jobs/:jobId.

Requires session authentication. Subject to rate limiting (per-IP) and workload gating (per-user). Messages longer than 800 characters consume 2 workload slots; shorter messages consume 1.

Request body

Field	Type	Required	Description
`message`	string	Yes	Message to send

{
  "message": "Summarize my tasks"
}

The sessionKey parameter has been removed. Chat sessions are now managed server-side by the backend job processor.

Response (202 Accepted)

{
  "success": true,
  "queued": true,
  "jobId": "job_abc123",
  "status": "queued",
  "agentId": "inst_a1b2c3d4e5",
  "agentName": "my-agent"
}

Field	Type	Description
`success`	boolean	Always `true` on success
`queued`	boolean	Always `true` — the message was enqueued
`jobId`	string	Job identifier. Poll `GET /api/jobs/:jobId` to retrieve the result.
`status`	string	`"queued"` — the job is waiting to be processed
`agentId`	string	The user’s OpenClaw instance identifier
`agentName`	string	Name of the agent

Errors

Code	Description
400	`Message required` — the `message` field is missing
401	`Unauthorized` — no valid session
404	`User not found` — authenticated user does not exist
404	`No OpenClaw instance found` — the user has no provisioned OpenClaw instance
404	`No agent found` — the user has no deployed agent
429	`Too many requests` — rate limit or workload gate exceeded. Response may include `retryAfterSeconds`.
500	`Chat failed` — error enqueuing the job
502	Backend job queue returned an error or non-JSON response
503	`No gateway token available` — no per-user gateway token was found in the database and no shared `OPENCLAW_GATEWAY_TOKEN` environment variable is configured

Example

curl -X POST https://agentbot.sh/api/gateway/chat \
  -H "Content-Type: application/json" \
  -H "Cookie: next-auth.session-token=YOUR_SESSION" \
  -d '{"message": "What is my agent status?"}'

Gateway status

GET /api/gateway/status

Requires session authentication. Returns a combined view of gateway health, active sessions, and cron jobs scoped to the authenticated user’s gateway. The server resolves the user’s per-user gateway URL and token from the database before invoking gateway tools. Use this endpoint to get a real-time snapshot of the gateway’s operational state.

Response

{
  "health": "healthy",
  "healthDetail": {
    "ok": true,
    "status": "healthy"
  },
  "sessions": {
    "total": 5,
    "active": 3,
    "list": [
      {
        "sessionKey": "main",
        "status": "active",
        "messageCount": 24,
        "lastActivity": "2026-03-30T01:15:00Z"
      }
    ]
  },
  "cron": {
    "total": 2,
    "enabled": 1,
    "jobs": [
      {
        "id": "heartbeat",
        "name": "Heartbeat",
        "enabled": true,
        "schedule": { "kind": "every", "everyMs": 1800000 },
        "lastRun": "2026-03-30T01:00:00Z",
        "nextRun": "2026-03-30T01:30:00Z"
      }
    ]
  }
}

Field	Type	Description
`health`	string	Overall gateway health: `healthy` or `unreachable`
`healthDetail`	object	Raw health check result. Contains `ok` (boolean) and `status` or `error` fields.
`sessions.total`	number	Total number of sessions on the gateway
`sessions.active`	number	Number of sessions with recent activity
`sessions.list`	array	Up to 10 recent sessions
`cron.total`	number	Total number of cron jobs
`cron.enabled`	number	Number of enabled cron jobs
`cron.jobs`	array	Up to 10 cron jobs

Errors

Code	Description
401	Unauthorized — no valid session

Example

curl -X GET https://agentbot.sh/api/gateway/status \
  -H "Cookie: next-auth.session-token=YOUR_SESSION"

Production gateway service

The production gateway is a modular Express application deployed on Railway that manages the OpenClaw gateway process. It provides a setup UI for initial configuration, an admin dashboard for monitoring, device pairing, and persistent storage on a Railway volume at /data. The wrapper proxies HTTP and WebSocket traffic to the internal OpenClaw gateway on 127.0.0.1:18789.

Wrapper authentication

Most wrapper management endpoints require admin authentication. When the WRAPPER_ADMIN_PASSWORD environment variable is set, requests must include one of:

Cookie: ocw_admin cookie set via the /login page (browser sessions)
Bearer token: Authorization: Bearer <password> header (API calls)
Legacy header: x-admin-token: <password> header

When WRAPPER_ADMIN_PASSWORD is not set, all management endpoints are accessible without authentication. Browser requests without valid credentials are redirected to /login. API requests receive a 401 JSON response.

Wrapper endpoints

The wrapper exposes the following management endpoints on its public port. All other requests are proxied to the internal OpenClaw gateway.

Gateway status

GET /api/status

No authentication required. Returns gateway state, uptime, and configuration status. Railway uses this endpoint as the container health check.

Response

{
  "state": "running",
  "running": true,
  "configured": true,
  "uptime": 3600.5,
  "terminalSessions": 2,
  "gatewayToken": "abc123...",
  "aiProvider": "google/gemini-2.5-flash",
  "runtime": {
    "ffmpeg": {
      "available": true,
      "version": "6.1.1"
    }
  },
  "ts": "2026-04-05T12:00:00.000Z"
}

Field	Type	Description
`state`	string	Gateway process state: `running`, `starting`, `stopped`, or `crashed`
`running`	boolean	`true` when the gateway process is running
`configured`	boolean	Always `true`
`uptime`	number	Wrapper process uptime in seconds
`terminalSessions`	number	Number of active WebSocket terminal sessions. Returns `0` when the terminal feature is unavailable. See WebSocket proxy.
`gatewayToken`	string \| null	The configured gateway authentication token, or `null` if not set
`aiProvider`	string \| null	The primary model configured in the gateway, or `null` if not configured
`runtime`	object	Runtime capabilities of the managed container
`runtime.ffmpeg.available`	boolean	Whether `ffmpeg` is installed and accessible in the container. Required for autonomous baseFM DJ broadcasting.
`runtime.ffmpeg.version`	string \| null	The ffmpeg version string, or `null` when ffmpeg is not available
`ts`	string	ISO 8601 timestamp of the status check

The previous online and logsCount fields have been replaced. Use running instead of online and GET /api/logs to retrieve log data.

Gateway logs

GET /api/logs

Requires admin authentication. Returns the last N lines of gateway process output.

Query parameters

Parameter	Type	Default	Description
`lines`	number	`100`	Number of log lines to return. Maximum `500`.

Response

{
  "logs": [
    {
      "ts": "2026-04-05T12:00:00.000Z",
      "stream": "stdout",
      "line": "Gateway ready on port 18789"
    }
  ]
}

Field	Type	Description
`logs`	array	Array of log entry objects
`logs[].ts`	string	ISO 8601 timestamp of the log entry
`logs[].stream`	string	Output stream: `stdout` or `stderr`
`logs[].line`	string	Log line content

Live log stream

GET /api/logs/stream

Requires admin authentication. Server-Sent Events (SSE) stream of gateway log output. On connection, the last 50 log entries are sent as history, followed by real-time log entries as they occur.

Event format

Each event is a JSON object with the same shape as entries in GET /api/logs:

{
  "ts": "2026-04-05T12:00:00.000Z",
  "stream": "stdout",
  "line": "Gateway ready on port 18789"
}

Gateway state events

GET /api/events

Requires admin authentication. Server-Sent Events (SSE) stream of gateway state changes. On connection, the current state is sent immediately, followed by real-time state change events.

Event format

{
  "state": "running"
}

Field	Type	Description
`state`	string	Gateway process state: `running`, `starting`, `stopped`, or `crashed`

Restart gateway

POST /api/gateway/restart

Requires admin authentication. Stops the gateway process and restarts it.

The endpoint path has changed from POST /api/restart to POST /api/gateway/restart.

Response (200)

{
  "ok": true,
  "message": "Gateway restarting..."
}

Error response (500)

{
  "ok": false,
  "error": "Restart failed: gateway process not found"
}

Stop gateway

POST /api/gateway/stop

Requires admin authentication. Stops the gateway process without restarting it.

Response (200)

{
  "ok": true,
  "message": "Gateway stopped."
}

Error response (500)

{
  "ok": false,
  "error": "Stop failed: ..."
}

Read gateway configuration

GET /api/config

Requires admin authentication. Returns the current openclaw.json configuration with sensitive fields redacted.

Response (200)

{
  "ok": true,
  "config": {
    "agents": {
      "defaults": {
        "model": { "primary": "google/gemini-2.5-flash" },
        "workspace": "/data/.openclaw/workspace"
      }
    },
    "gateway": {
      "mode": "local",
      "bind": "loopback",
      "port": 18789
    }
  }
}

Sensitive fields (botToken, token, appToken, apiKey, password, serviceAccount, secret, key, auth) are replaced with [redacted] in the response.

Errors

Code	Description
401	Unauthorized
404	No configuration file found

Update gateway configuration

POST /api/config

Requires admin authentication. Writes the provided JSON object as the new openclaw.json configuration. The gateway hot-reloads the configuration automatically.

Request body

The full configuration object to write. Must be a JSON object.

Response (200)

{
  "ok": true,
  "message": "Config written. Hot reload in progress..."
}

Errors

Code	Description
400	Invalid config body (not an object)
401	Unauthorized
500	Failed to write configuration

Device pairing

The wrapper provides endpoints for managing device pairing requests. All pairing endpoints require admin authentication.

List pending pairing requests

GET /api/pairing/pending

Response (200)

{
  "ok": true,
  "pending": [
    {
      "id": "req_abc123",
      "name": "Chrome on macOS",
      "ip": "192.168.1.42",
      "role": "viewer",
      "createdAt": "2026-04-05T12:00:00.000Z",
      "expiresAt": "2026-04-05T12:10:00.000Z"
    }
  ]
}

Field	Type	Description
`ok`	boolean	Always `true` on success
`pending`	array	Array of pending pairing request objects
`pending[].id`	string	Unique request identifier. Also accessible as `requestId`.
`pending[].name`	string	Device display name. Also accessible as `deviceName`.
`pending[].ip`	string	Remote IP address of the requesting device. Also accessible as `remoteIp`.
`pending[].role`	string	Requested access role (for example `viewer`, `admin`)
`pending[].createdAt`	string	ISO 8601 timestamp of when the request was created
`pending[].expiresAt`	string \| undefined	ISO 8601 timestamp of when the request expires. When present, the admin dashboard displays a countdown timer.

Approve a pairing request

POST /api/pairing/approve

Request body

Field	Type	Required	Description
`requestId`	string	Yes	The pairing request ID to approve

Response (200)

{
  "ok": true,
  "message": "Request abc123 approved"
}

Reject a pairing request

POST /api/pairing/reject

Request body

Field	Type	Required	Description
`requestId`	string	Yes	The pairing request ID to reject

Response (200)

{
  "ok": true,
  "message": "Request abc123 rejected"
}

Revoke a paired device

POST /api/pairing/revoke

Request body

Field	Type	Required	Description
`deviceId`	string	Yes	The device ID to revoke
`role`	string	Yes	The device role to revoke

Response (200)

{
  "ok": true,
  "message": "Device abc123 revoked"
}

List paired devices

GET /api/pairing/paired

Response (200)

{
  "ok": true,
  "paired": [
    {
      "id": "dev_abc123",
      "name": "Chrome on macOS",
      "role": "viewer",
      "pairedAt": "2026-04-05T12:01:00.000Z"
    }
  ]
}

Field	Type	Description
`ok`	boolean	Always `true` on success
`paired`	array	Array of paired device objects
`paired[].id`	string	Unique device identifier. Also accessible as `deviceId`.
`paired[].name`	string	Device display name. Also accessible as `deviceName`.
`paired[].role`	string	Access role assigned to the device
`paired[].pairedAt`	string	ISO 8601 timestamp of when the device was paired

Pairing event stream

GET /api/pairing/stream

Requires admin authentication. Server-Sent Events (SSE) stream of real-time pairing updates. On connection, the current list of pending requests is sent immediately, followed by change events as they occur.

Event format

Pending list update — sent on connection and whenever the pending list changes (for example when a new pairing request arrives):

{
  "type": "pending",
  "pending": [
    {
      "id": "req_abc123",
      "name": "Chrome on macOS",
      "ip": "192.168.1.42",
      "role": "viewer",
      "createdAt": "2026-04-05T12:00:00.000Z",
      "expiresAt": "2026-04-05T12:10:00.000Z"
    }
  ]
}

Pairing action update — sent when a pairing request is approved, rejected, or a paired device is revoked:

{
  "type": "update",
  "action": "approved",
  "requestId": "req_abc123"
}

Field	Type	Description
`type`	string	Event type: `pending` or `update`
`pending`	array	Full list of pending requests. Only present when `type` is `pending`. See list pending pairing requests for the object shape.
`action`	string	Action that occurred: `approved`, `rejected`, or `revoked`. Only present when `type` is `update`.
`requestId`	string	The pairing request ID that was approved or rejected. Present when `action` is `approved` or `rejected`.
`deviceId`	string	The device ID that was revoked. Present when `action` is `revoked`.

Setup flow

The wrapper provides a setup UI for initial gateway configuration. When the gateway is not yet configured, requests to / are redirected to /setup.

Setup page

GET /setup

No authentication required. Serves the setup UI HTML. Redirects to / if the gateway is already running.

Save setup configuration

POST /setup/save

No authentication required. Validates the setup form, writes the configuration and environment files, and launches the gateway.

Request body

Field	Type	Required	Description
`provider`	string	Yes	Model provider: `anthropic`, `openai`, `google`, `openrouter`, `groq`, `moonshot`, `zai`, `minimax`, or `ollama`
`apiKey`	string	Conditional	API key for the selected provider. Required for all providers except `ollama`.
`ollamaUrl`	string	Conditional	Ollama base URL (e.g., `http://localhost:11434`). Required when `provider` is `ollama`.
`model`	string	Conditional	Model to use. Required for `ollama`. Optional for other providers (defaults to provider-specific model).
`fallbackModel`	string	No	Fallback model ID
`telegramBotToken`	string	No	Telegram bot token
`telegramDmPolicy`	string	No	Telegram DM policy: `pairing`, `allowlist`, `open`, or `disabled`. Defaults to `pairing`.
`telegramAllowFrom`	string	No	Comma-separated list of allowed Telegram users
`telegramWebhookUrl`	string	No	Telegram webhook URL
`discordBotToken`	string	No	Discord bot token
`discordDmPolicy`	string	No	Discord DM policy: `pairing`, `allowlist`, `open`, or `disabled`. Defaults to `pairing`.
`discordAllowFrom`	string	No	Comma-separated list of allowed Discord users
`slackBotToken`	string	No	Slack bot token (requires `slackAppToken`)
`slackAppToken`	string	No	Slack app token (requires `slackBotToken`)
`slackDmPolicy`	string	No	Slack DM policy: `pairing`, `allowlist`, `open`, or `disabled`. Defaults to `pairing`.
`mattermostUrl`	string	No	Mattermost server URL (requires `mattermostToken`)
`mattermostToken`	string	No	Mattermost token (requires `mattermostUrl`)
`mattermostTeam`	string	No	Mattermost team name
`sessionScope`	string	No	Session scope: `main`, `per-peer`, `per-channel-peer`, or `per-account-channel-peer`. Defaults to `per-channel-peer`.
`sessionResetMode`	string	No	Session reset mode: `off`, `daily`, or `idle`
`sessionResetHour`	string	No	Hour (UTC) for daily session reset

Response (200)

{
  "ok": true,
  "message": "Config saved. Gateway launching..."
}

Errors

Code	Description
400	Validation errors. Response includes an `errors` array with descriptive messages.
409	Gateway is already running or starting. Use `POST /api/config` to update configuration instead.
500	Failed to write configuration or launch gateway

Get Ollama configuration

GET /setup/api/ollama-config

No authentication required. Returns the pre-configured Ollama base URL from environment variables.

Response

{
  "ollamaBaseUrl": "http://ollama.railway.internal:11434"
}

List Ollama models

GET /setup/api/ollama-models

No authentication required. Proxies to an Ollama instance to fetch available models.

Query parameters

Parameter	Type	Required	Description
`url`	string	No	Ollama base URL. Falls back to the `OLLAMA_BASE_URL` environment variable if not provided.

Response (200)

{
  "models": ["llama3.3", "deepseek-r1:1.5b"]
}

Errors

Code	Description
400	No Ollama URL provided or invalid URL format
502	Could not reach Ollama instance or request timed out

Approve channel pairing via setup

POST /setup/pairing/approve

No authentication required. Approves a channel pairing code during initial setup.

Request body

Field	Type	Required	Description
`channel`	string	Yes	Channel name
`code`	string	Yes	Pairing code

Response (200)

{
  "ok": true,
  "output": "Pairing approved"
}

Reset gateway

POST /setup/reset

No authentication required. Stops the gateway and resets configuration.

Request body

Field	Type	Required	Description
`mode`	string	No	Reset mode: omit for config-only reset, or `full` for factory reset (wipes all data)

Response (200)

Config-only reset:

{
  "ok": true,
  "message": "Config reset complete. Redirecting to setup..."
}

Full factory reset:

{
  "ok": true,
  "message": "Full reset complete. All data wiped. Redirecting to setup..."
}

Export gateway data

GET /setup/export

No authentication required. Downloads a zip archive of all gateway data. When WRAPPER_ADMIN_PASSWORD is set, the zip is password-protected with the admin password.

Response

Returns a application/zip file as a download attachment. The Content-Disposition header includes a filename in the format openclaw-export-<timestamp>.zip.

Errors

Code	Description
404	No data directory found to export
500	Failed to create export archive

Proxy error response (503)

When a request is proxied to the gateway but the gateway is not running, the wrapper returns a 502 error:

{
  "error": "Gateway not reachable",
  "detail": "connect ECONNREFUSED 127.0.0.1:18789"
}

WebSocket proxy

The wrapper handles WebSocket upgrade requests with path-based routing:

/ws/terminal — Routed to the built-in terminal WebSocket service, which provides an interactive terminal session to the container. This feature is optional — it requires node-pty to be installed in the container. When node-pty is not available, the terminal service is disabled and connections to /ws/terminal receive an error message indicating the feature is unavailable. The terminalSessions field in GET /api/status returns 0 when the terminal is disabled.
All other paths — Proxied to the internal OpenClaw gateway. The wrapper injects the Authorization: Bearer <token> header automatically.

The node-pty native module is dynamically loaded at startup. If it cannot be compiled or is not present in the container image, the gateway logs a warning (node-pty not available — terminal feature disabled) and continues operating normally. All other gateway features remain functional. To enable the terminal, ensure node-pty is installed and its native bindings can compile for the target platform.

When the gateway is not running, non-terminal WebSocket upgrade requests are proxied to the gateway address, which returns a connection error.

Terminal WebSocket messages

Messages on the /ws/terminal WebSocket are JSON objects with a type field.

Client-to-server messages

Type	Fields	Description
`input`	`data` (string)	Send keystrokes or pasted text to the terminal
`resize`	`cols` (number), `rows` (number)	Notify the server that the terminal dimensions have changed

Server-to-client messages

Type	Fields	Description
`output`	`data` (string)	Terminal output to render
`exit`	`code` (number)	The terminal process exited with the given exit code

Environment variables

Variable	Default	Description
`OPENCLAW_GATEWAY_TOKEN`	(empty)	Bearer token for authenticating proxied requests to the internal gateway. When not set, a random 64-character hex token is generated at startup.
`PORT`	`3000`	Port the wrapper listens on
`OPENCLAW_DATA_DIR`	`/data`	Root directory for persistent storage. The OpenClaw configuration and workspace are stored under `{OPENCLAW_DATA_DIR}/.openclaw/`. This should be a Railway volume mount.
`WRAPPER_ADMIN_PASSWORD`	(empty)	Password for accessing the admin dashboard and protected API endpoints. When not set, all management endpoints are accessible without authentication. See wrapper authentication.
`OLLAMA_BASE_URL`	(empty)	Ollama base URL (e.g., `http://ollama.railway.internal:11434`). Pre-fills the Ollama URL field in the setup UI and is used as the default for the Ollama models endpoint.
`GEMINI_API_KEY`	(empty)	Google Gemini API key. Written into the gateway `.env` file at startup for model authentication.
`OPENROUTER_API_KEY`	(empty)	OpenRouter API key for LLM provider authentication. Written into the gateway `.env` file at startup.

OpenClaw configuration and secrets

The wrapper writes an openclaw.json configuration file to {OPENCLAW_DATA_DIR}/.openclaw/openclaw.json at startup. API keys are stored separately in {OPENCLAW_DATA_DIR}/.openclaw/.env to keep secrets out of the JSON configuration file. The gateway reads both files at startup.

`.env` variable	Description
`GEMINI_API_KEY`	Google Gemini API key for model authentication. Populated from the `GEMINI_API_KEY` container environment variable at startup.
`OPENROUTER_API_KEY`	OpenRouter API key for LLM provider authentication. Populated from the `OPENROUTER_API_KEY` container environment variable at startup.

Default model configuration

The production gateway configures the following model default:

Parameter	Value	Description
`agents.defaults.model.primary`	`google/gemini-2.5-flash`	Primary model for agent requests

When you use the setup UI to configure the gateway, the primary model is set based on your chosen provider. If you do not specify a model, the following defaults are used:

Provider	Default model
`anthropic`	`anthropic/claude-opus-4-6`
`openai`	`openai/gpt-5.4`
`google`	`google/gemini-2.5-pro`
`openrouter`	`openrouter/auto`
`groq`	`groq/llama-3.3-70b-versatile`
`moonshot`	`moonshot/kimi-k2.5`
`zai`	`zai/glm-4.5`
`minimax`	`minimax/MiniMax-M2.7`
`ollama`	(must be specified)

Agent defaults

Parameter	Value	Description
`agents.defaults.workspace`	`{OPENCLAW_DATA_DIR}/.openclaw/workspace`	Default workspace directory inside the persistent volume
`agents.defaults.userTimezone`	`Europe/London`	Default timezone
`agents.defaults.thinkingDefault`	`low`	Default thinking verbosity level
`agents.defaults.verboseDefault`	`off`	Default verbose output mode
`agents.defaults.timeoutSeconds`	`600`	Maximum time in seconds for a single request
`agents.defaults.maxConcurrent`	`3`	Maximum concurrent agent tasks
`agents.defaults.heartbeat.every`	`30m`	Self-monitoring heartbeat interval
`agents.defaults.heartbeat.lightContext`	`true`	Use minimal context for heartbeat checks
`agents.defaults.heartbeat.isolatedSession`	`true`	Run heartbeat in an isolated session to avoid polluting active conversations

Tool configuration

Parameter	Value	Description
`tools.profile`	`coding`	Tool profile loaded by default
`tools.exec.backgroundMs`	`10000`	Maximum time in milliseconds a background process can run
`tools.exec.timeoutSec`	`1800`	Maximum execution time in seconds for foreground commands (30 minutes)
`tools.web.search.enabled`	`true`	Web search tool is available
`tools.web.fetch.enabled`	`true`	Web fetch tool is available
`tools.web.fetch.maxChars`	`50000`	Maximum characters returned from web fetch requests

Session configuration

Parameter	Value	Description
`session.scope`	`per-sender`	Each sender gets an isolated conversation session
`session.reset.mode`	`daily`	Sessions reset on a daily schedule
`session.reset.atHour`	`4`	Hour (UTC) when daily session reset occurs
`session.maintenance.mode`	`warn`	Log warnings when sessions exceed size limits instead of force-pruning
`session.maintenance.pruneAfter`	`30d`	Inactive sessions are pruned after 30 days
`session.maintenance.maxEntries`	`500`	Maximum entries per session before maintenance triggers

Cron configuration

Parameter	Value	Description
`cron.enabled`	`true`	Scheduled task execution is enabled
`cron.maxConcurrentRuns`	`2`	Maximum number of cron jobs that can run simultaneously
`cron.sessionRetention`	`24h`	How long cron session data is retained after completion

Logging configuration

Parameter	Value	Description
`logging.level`	`info`	Minimum log level written to persistent storage
`logging.consoleLevel`	`info`	Minimum log level written to stdout
`logging.consoleStyle`	`compact`	Log output format (`compact` omits timestamps and metadata for cleaner output)

Gateway settings

Parameter	Value	Description
`gateway.mode`	`local`	Gateway operating mode
`gateway.bind`	`loopback`	Bind address. The internal gateway binds to loopback only; the wrapper handles external traffic and proxies it to `127.0.0.1:18789`.
`gateway.port`	`18789`	Internal gateway port
`gateway.auth.token`	Auto-generated or `OPENCLAW_GATEWAY_TOKEN`	Gateway authentication token. Written directly into the configuration when `OPENCLAW_GATEWAY_TOKEN` is set.
`gateway.trustedProxies`	`["127.0.0.1", "::1"]`	IP addresses trusted as reverse proxies. Includes IPv4 and IPv6 loopback only, since all external traffic passes through the local Express wrapper.
`gateway.controlUi.allowedOrigins`	`["*"]`	Permitted WebSocket origins. Set to `["*"]` to allow connections from any origin.
`gateway.controlUi.allowInsecureAuth`	`true`	Allows the gateway to accept token-based auth from the wrapper proxy without requiring device pairing for browser sessions.
`gateway.reload.mode`	`hybrid`	Configuration reload strategy

Health check

The container image is configured with a Docker HEALTHCHECK that probes GET /api/status every 15 seconds (5-second timeout, 60-second start period, 5 retries). Railway uses the same path (/api/status) for health monitoring and will restart the container on failure.

Persistent storage

The wrapper stores all OpenClaw state under {OPENCLAW_DATA_DIR}/.openclaw/ (default: /data/.openclaw/). This directory should be backed by a Railway volume mount so that configuration, conversations, and workspace files survive container restarts.

Process management

The wrapper manages the OpenClaw gateway process with automatic restart on crash. When the gateway exits unexpectedly, the wrapper schedules a restart with exponential backoff starting at 2 seconds, doubling each attempt, up to a maximum of 30 seconds. The gateway is considered ready when it accepts TCP connections on port 18789. If the gateway does not become ready within 60 seconds, the wrapper marks it as crashed and schedules another restart attempt.

Control UI URL resolution

The dashboard constructs control UI links (for the chat, skills, and config views) using the user’s own gateway URL as the base origin. When a user has a deployed gateway instance, the openclawUrl stored in their account is used to derive the control UI base URL. The platform default gateway URL is only used as a fallback when no user-specific URL is available.

URL construction

The control UI URL is built from the following components:

Component	Source	Description
Base origin	User’s `openclawUrl`	The origin (scheme + host) of the user’s deployed gateway. Falls back to the platform default `OPENCLAW_GATEWAY_URL` when not set.
View path	`/chat`, `/skills`, or `/config`	The control UI view to open
Query parameters	View-specific	For the chat view, includes `session` (defaults to `main`)
Hash fragment	Gateway credentials	Contains `token` (gateway auth token) and `gatewayUrl` (WebSocket URL). Passed in the URL fragment so they are never sent to the server.

Example

For a user with openclawUrl set to https://my-agent.up.railway.app:

https://my-agent.up.railway.app/chat?session=main#token=abc123&gatewayUrl=wss%3A%2F%2Fmy-agent.up.railway.app

When openclawUrl is not set, the platform default gateway URL is used instead:

https://gateway.agentbot.sh/chat?session=main#token=abc123&gatewayUrl=wss%3A%2F%2Fgateway.agentbot.sh

The gateway token and WebSocket URL are passed in the URL hash fragment, which is never sent to the server in HTTP requests. This ensures credentials remain client-side only.

WebSocket URL derivation

The WebSocket gateway URL is derived from the user’s gateway URL by replacing the scheme with wss:// and using the same host. For example, https://my-agent.up.railway.app becomes wss://my-agent.up.railway.app.

Per-agent gateway authentication

Each agent container receives a unique gateway auth token at provisioning time. The internal gateway authenticates requests using token-based auth on port 18789.

Field	Description
`gateway.auth.mode`	`token`
`gateway.auth.token`	Unique hex token auto-generated per container. The container manager and Railway provisioning paths generate a 64-character token (32 bytes of entropy). The backend provisioning path generates a 48-character token (24 bytes of entropy). A 64-character token (32 bytes) is generated at startup if no token is passed via the `OPENCLAW_GATEWAY_TOKEN` environment variable.
`gateway.port`	`18789`

Token resolution order

When the platform sends requests to a user’s gateway (for example, the gateway chat proxy or gateway status endpoints), the server resolves the authentication token in the following order:

Per-user gateway token — the server looks up the authenticated user’s gateway URL from the database. If a per-user URL is found, the token associated with that user’s gateway is used.
Shared gateway token — if no per-user gateway is found, the server falls back to the OPENCLAW_GATEWAY_TOKEN environment variable.

If neither source provides a token, the request fails with a 503 error. This resolution order ensures that each user’s gateway traffic is authenticated with their own token rather than a single shared credential.

The container startup process writes its own minimal configuration to $HOME/.openclaw/openclaw.json using a slightly different schema (auth.method at the top level instead of gateway.auth.mode). The provisioning config written by the backend uses the gateway.auth.mode path. When the container starts, it overwrites the provisioning config with its own minimal skeleton. All provisioning paths include a top-level env section containing OPENROUTER_API_KEY so the OpenClaw runtime can authenticate with the LLM provider. To preserve the full provisioning config, pass the gateway token via the OPENCLAW_GATEWAY_TOKEN environment variable so the startup process uses the same token.

Agent container configuration

When an agent is provisioned, the backend generates an OpenClaw configuration with the following parameters. These values are set automatically and cannot be overridden by the caller.

OpenClaw configuration `env` section

Each agent container’s openclaw.json includes a top-level env section that passes secrets to the OpenClaw runtime. All three provisioning paths (production gateway startup, Railway direct provisioning, and backend container manager) write this section into the configuration file at launch.

Parameter	Description
`env.OPENROUTER_API_KEY`	OpenRouter API key for LLM provider authentication. Populated from the `OPENROUTER_API_KEY` container environment variable. Required for the agent to make model requests through OpenRouter.

The env section in openclaw.json is distinct from the container-level environment variables listed below. Container environment variables are set on the container process by the orchestrator (Railway or Docker). The env section is read by the OpenClaw runtime from its configuration file and used internally for service authentication. Both mechanisms deliver the same OPENROUTER_API_KEY value, but the config-file path ensures OpenClaw can access the key even when the runtime does not inherit the container’s full environment.

Container environment variables

The following environment variables are set on every agent container at launch. Variables are grouped by source — some are set by the container startup process (local Docker path) and others are injected by the provisioning service (Railway path). When both paths set the same variable, the provisioning service value takes precedence.

Variable	Default	Description
`NODE_ENV`	`production`	Node.js environment. Set by the provisioning service.
`PORT`	Railway-assigned	Port the agent process listens on inside the container. This value is injected by Railway automatically and is no longer set by the provisioning service. A TCP proxy inside the container forwards traffic from the Railway-assigned port to the OpenClaw gateway on `127.0.0.1:18789`.
`HOME`	`/root`	Home directory for the container user. Managed runtime containers run as `root` with the OpenClaw data directory at `/root/.openclaw`.
`TERM`	`xterm-256color`	Terminal type
`NODE_COMPILE_CACHE`	`/var/tmp/openclaw-compile-cache`	Directory for the Node.js compile cache. Speeds up cold starts by caching compiled bytecode across process restarts.
`OPENCLAW_NO_RESPAWN`	`1`	When set to `1`, prevents the OpenClaw process from automatically respawning after exit. Container-level restart policies (Docker `--restart` or orchestrator health checks) handle process recovery instead.
`OPENCLAW_GATEWAY_TOKEN`	Auto-generated	Gateway authentication token. When not set, a 64-character hex token (32 bytes of entropy) is generated at startup. See per-agent gateway authentication.
`OPENCLAW_GATEWAY_URL`	`http://openclaw-gateway-lqma:10000`	Gateway URL for the agent to communicate with the platform gateway. Set by the provisioning service. When this variable is not configured, the provisioning endpoint falls back to the default internal Railway DNS address shown here. You can override this value to point to a custom gateway deployment.
`OPENCLAW_GATEWAY_PORT`	`18789`	Port the gateway listens on inside the container
`OPENCLAW_GATEWAY_BIND`	—	Deprecated. Previously set to `0.0.0.0` to allow Railway’s reverse proxy to route traffic into the container. This variable is no longer injected by the provisioning service because it has no effect on OpenClaw. Traffic now reaches the gateway through a TCP proxy that forwards the Railway-assigned `PORT` to `127.0.0.1:18789`.
`OPENCLAW_BIND`	—	Deprecated. Previously used as a fallback bind address when `OPENCLAW_GATEWAY_BIND` was not set. This variable is no longer injected by the provisioning service.
`AGENTBOT_USER_ID`	Per-user	Owner user ID passed at provisioning time
`AGENTBOT_PLAN`	`solo`	Subscription plan tier (`solo`, `collective`, `label`, or `network`)
`AGENTBOT_API_URL`	`https://agentbot-backend-production.up.railway.app`	Backend API URL for the agent to call platform services. Set by the provisioning service.
`AGENTBOT_MODE`	`home`	Installation mode (`home` for self-hosted, `link` for existing OpenClaw)
`AGENTBOT_API_KEY`	Per-user	API key for authenticating with the Agentbot platform
`DATABASE_URL`	Platform-provided	PostgreSQL connection string for agent data persistence. Set by the provisioning service.
`OPENROUTER_API_KEY`	Platform-provided	OpenRouter API key for AI model access. Set by the provisioning service.
`INTERNAL_API_KEY`	Platform-provided	Internal API key for authenticating agent-to-platform requests. Set by the provisioning service.
`WALLET_ENCRYPTION_KEY`	Platform-provided	Encryption key for securing agent wallet data at rest. Set by the provisioning service.
`CONTROL_UI_ORIGIN`	(empty)	Primary origin allowed to connect to the Control UI via WebSocket. When not set, falls back to `NEXT_PUBLIC_APP_URL`, then `https://agentbot.sh`.
`CONTROL_UI_COMPAT_ORIGIN`	(empty)	Optional secondary origin for Control UI WebSocket connections. When set, this origin is added alongside the primary origin in the `controlUi.allowedOrigins` list.

Gateway settings

Parameter	Value	Description
`gateway.mode`	`local`	Gateway operating mode
`gateway.bind`	`lan`	Bind address for the gateway. Set to `lan` so the gateway listens on all interfaces inside the Docker container, which is required for port-mapped containers where the host forwards traffic to the container’s published port. On Railway-provisioned containers, a TCP proxy may still forward traffic from the Railway-assigned `PORT` to `127.0.0.1:18789`.
`gateway.port`	`18789`	Internal gateway port
`gateway.auth.mode`	`token`	Authentication mode
`gateway.auth.rateLimit.maxAttempts`	`10`	Maximum authentication attempts before lockout
`gateway.auth.rateLimit.windowMs`	`60000`	Rate limit window in milliseconds (1 minute)
`gateway.auth.rateLimit.lockoutMs`	`300000`	Lockout duration in milliseconds (5 minutes)
`gateway.auth.rateLimit.exemptLoopback`	`true`	Exempt loopback addresses from rate limiting
`gateway.auth.allowTailscale`	`true`	Allow Tailscale network connections
`gateway.trustedProxies`	`["127.0.0.1", "10.0.0.0/8", "100.64.0.0/10", "172.16.0.0/12", "192.168.0.0/16"]`	IP addresses trusted as reverse proxies. Requests from these addresses have their `X-Forwarded-For` headers honored for origin detection. Includes loopback, RFC 1918 private ranges, and the `100.64.0.0/10` CGNAT range used by Railway’s internal network.
`gateway.http.endpoints.chatCompletions.enabled`	`true`	Enables the OpenAI-compatible `POST /v1/chat/completions` endpoint on the gateway. Required for the REST-based `POST /api/chat` endpoint. Set during provisioning.
`gateway.controlUi.enabled`	`true`	Gateway control UI is enabled in containers.
`gateway.controlUi.allowedOrigins`	Derived from environment	Origins permitted to connect to the Control UI via WebSocket. Defaults to the value of `CONTROL_UI_ORIGIN` (or `NEXT_PUBLIC_APP_URL`, falling back to `https://agentbot.sh`). An optional second origin can be added via `CONTROL_UI_COMPAT_ORIGIN`. When not configured, the gateway may reject WebSocket connections with an “origin not allowed” error if the request origin does not match.
`gateway.controlUi.dangerouslyDisableDeviceAuth`	`false`	When `true`, disables the interactive device authentication flow for the Control UI. Defaults to `false` — device auth is now required. For headless deployments where no browser is available, you may need to use token-based authentication instead of the device auth handshake.
`gateway.controlUi.dangerouslyAllowHostHeaderOriginFallback`	`false`	When `true`, allows the gateway to fall back to the `Host` header when the `Origin` header is missing from a WebSocket upgrade request. Defaults to `false` — the `Origin` header is now required.

The controlUi.allowedOrigins, controlUi.dangerouslyDisableDeviceAuth, and controlUi.dangerouslyAllowHostHeaderOriginFallback defaults have changed. Previously, allowedOrigins was set to ["*"], and both dangerouslyDisableDeviceAuth and dangerouslyAllowHostHeaderOriginFallback were true. The new defaults restrict origins to the platform URL and require device authentication and the Origin header for WebSocket connections.

Tool settings

Parameter	Value	Description
`tools.profile`	`messaging` or `coding`	`messaging` for solo plan, `coding` for all other plans
`tools.deny`	`["browser", "canvas"]`	Tools disabled in container environments
`tools.exec.allowedCommands`	Array of whitelisted commands	Commands the agent is permitted to execute (includes `git`, `node`, `npm`, `python3`, `curl`, `ls`, `cat`, `grep`, `find`, `wget`, `mkdir`, `cp`, `mv`, `rm`, `echo`, `date`, `whoami`, `chmod`, `chown`, `touch`, `head`, `tail`, `wc`, `sort`, `uniq`, `awk`, `sed`, `tar`, `zip`, `unzip`, `docker`, `ps`, `df`, `du`)
`tools.exec.allowedPaths`	`["/root/.openclaw/workspace", "/tmp", "/root"]`	Filesystem paths the agent can access. Managed runtime containers run as `root` with configuration stored at `/root/.openclaw/openclaw.json`.
`tools.exec.denyPaths`	`["/etc/shadow", "/etc/passwd", "/proc", "/sys"]`	Filesystem paths the agent is blocked from accessing
`tools.web.maxChars`	`50000`	Maximum characters returned from web tool requests
`tools.loopDetection.maxIterations`	`20`	Maximum loop iterations before the agent is interrupted
`tools.loopDetection.windowMinutes`	`5`	Time window for loop detection

Session settings

Parameter	Value	Description
`session.maxTokens`	`100000`	Maximum tokens per session
`session.compaction.strategy`	`auto`	Automatic context compaction strategy
`session.compaction.triggerAtPercent`	`80`	Compaction triggers when token usage reaches this percentage

Agent defaults

Parameter	Value	Description
`agents.defaults.workspace`	`/root/.openclaw/workspace`	Default workspace directory. Managed runtime containers mount a persistent volume at `/root/.openclaw`.
`agents.defaults.imageMaxDimensionPx`	`1200`	Maximum image dimension in pixels (optimizes vision token usage)
`agents.defaults.userTimezone`	`Europe/London`	Default timezone (overridden by signup timezone when available)
`agents.defaults.timeFormat`	`24h`	Time format
`agents.defaults.compaction.maxMessages`	`200`	Maximum messages before compaction
`agents.defaults.compaction.keepLastN`	`20`	Number of recent messages preserved after compaction
`agents.defaults.heartbeat.every`	`30m`	Self-monitoring heartbeat interval
`agents.defaults.skipBootstrap`	`false`	Whether to skip the bootstrap phase
`agents.defaults.bootstrapMaxChars`	`4000`	Maximum characters for bootstrap content

Health monitoring

The gateway monitors channel health for each agent container. When a channel becomes unresponsive, the gateway can automatically restart it.

Parameter	Value	Description
`channelHealthCheckMinutes`	`5`	Interval between health checks for each channel
`channelStaleEventThresholdMinutes`	`30`	Channel is considered stale if no events are received within this window
`channelMaxRestartsPerHour`	`10`	Maximum number of automatic channel restarts per hour

CORS

The gateway supports CORS preflight via OPTIONS /api/v1/gateway. Allowed methods are GET, POST, and OPTIONS. The Content-Type, Authorization, X-Plugin-Id, and Payment headers are permitted in the CORS configuration.

The X-Payment-Method, X-Session-Id, and X-Wallet-Address headers are read server-side but are not included in the CORS Access-Control-Allow-Headers response. Cross-origin requests that include these headers may be rejected by the browser preflight check. Same-origin requests are unaffected. If you need to send these headers from a different origin, configure the GATEWAY_CORS_ORIGIN environment variable or proxy the request through a same-origin endpoint.

OpenAI-compatible endpoints

The gateway exposes a set of endpoints that follow the OpenAI API format. These allow you to use Agentbot as a drop-in replacement for the OpenAI SDK or any tool that supports the OpenAI API shape.

List models

GET /v1/models

Returns all available models in OpenAI-compatible format. This endpoint is public and does not require authentication.

Response (200)

{
  "object": "list",
  "data": [
    {
      "id": "openrouter/openai/gpt-4o-mini",
      "object": "model",
      "created": 1742472000,
      "owned_by": "openrouter"
    }
  ]
}

Model object

Field	Type	Description
`id`	string	Model identifier (provider-prefixed)
`object`	string	Always `model`
`created`	number	Unix timestamp of when the response was generated
`owned_by`	string	Provider that serves the model (e.g., `openrouter`, `agentbot`)

Errors

Status	Description
500	`Failed to fetch models` — an internal error occurred while retrieving the model list

Retrieve a model

GET /v1/models/:model

Returns details for a single model by its ID. This endpoint is public and does not require authentication.

Path parameters

Parameter	Type	Required	Description
`model`	string	Yes	The model ID to look up (e.g., `openrouter/openai/gpt-4o-mini`)

Response (200)

{
  "id": "openrouter/openai/gpt-4o-mini",
  "object": "model",
  "created": 1742472000,
  "owned_by": "openrouter"
}

Errors

Status	Description
404	`Model {model} not found` — no model matches the provided ID
500	`Failed to fetch model` — an internal error occurred

Create embeddings

POST /v1/embeddings

Generates embeddings for the given input. Proxies the request to OpenRouter. Requires authentication.

Headers

Header	Type	Required	Description
`Authorization`	string	Yes	Bearer token or session cookie
`Content-Type`	string	Yes	Must be `application/json`

Request body

{
  "input": "The quick brown fox jumps over the lazy dog",
  "model": "openai/text-embedding-3-small"
}

Field	Type	Required	Description
`input`	string or string[]	Yes	Text to generate embeddings for. Can be a single string or an array of strings.
`model`	string	No	Embedding model to use. Defaults to `openai/text-embedding-3-small`.

Response (200)

The response follows the OpenAI embeddings format. The exact shape depends on the upstream provider.

{
  "object": "list",
  "data": [
    {
      "object": "embedding",
      "index": 0,
      "embedding": [0.0023064255, -0.009327292, ...]
    }
  ],
  "model": "openai/text-embedding-3-small",
  "usage": {
    "prompt_tokens": 9,
    "total_tokens": 9
  }
}

Errors

Status	Description
400	`input is required` — the `input` field is missing from the request body
401	Unauthorized — valid authentication is required
503	`Embeddings not configured` — the server does not have an OpenRouter API key configured
500	`Embeddings request failed` — the upstream provider returned an error

Rate limits

Endpoint	Limit
`/api/v1/gateway`	100/min
`/v1/models`	120/min (general)
`/v1/models/:model`	120/min (general)
`/v1/embeddings`	120/min (general)

Examples

Route a request to the agent plugin (Stripe)

curl -X POST https://agentbot.sh/api/v1/gateway \
  -H "Content-Type: application/json" \
  -H "X-Plugin-Id: agent" \
  -H "Cookie: next-auth.session-token=YOUR_SESSION" \
  -d '{"messages": [{"role": "user", "content": "Summarize my tasks"}]}'

Route a request with MPP payment

curl -X POST https://agentbot.sh/api/v1/gateway \
  -H "Content-Type: application/json" \
  -H "X-Plugin-Id: generate-text" \
  -H "Authorization: Payment {\"scheme\":\"Payment\",\"transaction\":\"0x76...\",\"challengeNonce\":\"abc123\"}" \
  -d '{"messages": [{"role": "user", "content": "Hello"}]}'

Route a request with session-based billing

curl -X POST https://agentbot.sh/api/v1/gateway \
  -H "Content-Type: application/json" \
  -H "X-Plugin-Id: agent" \
  -H "X-Payment-Method: session" \
  -H "X-Session-Id: ses_a1b2c3d4e5f6..." \
  -H "X-Wallet-Address: 0xd8fd0e1dce89beaab924ac68098ddb17613db56f" \
  -d '{"messages": [{"role": "user", "content": "Summarize my tasks"}]}'

The response includes the remaining session balance:

Payment-Receipt: session:ses_a1b2c3d4e5f6...:v_abc123def456
X-Session-Remaining: 9.95

List available models (OpenAI-compatible)

curl https://agentbot.sh/v1/models

Retrieve a specific model

curl https://agentbot.sh/v1/models/openrouter/openai/gpt-4o-mini

Generate embeddings

curl -X POST https://agentbot.sh/v1/embeddings \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -d '{"input": "Hello world", "model": "openai/text-embedding-3-small"}'

Getting Started

Core Concepts

Integrations

Core Services

Payments

API Reference

MCP

Security

Documentation Index

​Gateway

​Base URL

​Request

​Headers

​Body

​Plugins

​Authentication

​Payment flow

​MPP 402 challenge

​Session-based billing

​Response

​Success

​Response headers

​Error responses

​Gateway chat proxy

​Request body

​Response (202 Accepted)

​Errors

​Example

​Gateway status

​Response

​Errors

​Example

​Production gateway service

​Wrapper authentication

​Wrapper endpoints

​Gateway status

Response

​Gateway logs

Query parameters

Response

​Live log stream

Event format

​Gateway state events

Event format

​Restart gateway

Response (200)

Error response (500)

​Stop gateway

Response (200)

Error response (500)

​Read gateway configuration

Response (200)

Errors

​Update gateway configuration

Request body

Response (200)

Errors

​Device pairing

​List pending pairing requests

Response (200)

​Approve a pairing request

Request body

Response (200)

​Reject a pairing request

Request body

Response (200)

​Revoke a paired device

Request body

Response (200)

​List paired devices

Response (200)

​Pairing event stream

Event format

​Setup flow

​Setup page

​Save setup configuration

Request body

Response (200)

Errors

​Get Ollama configuration

Gateway

Base URL

Request

Headers

Body

Plugins

Authentication

Payment flow

MPP 402 challenge

Session-based billing

Response

Success

Response headers

Error responses

Gateway chat proxy

Request body

Response (202 Accepted)

Errors

Example

Gateway status

Response

Errors

Example

Production gateway service

Wrapper authentication

Wrapper endpoints

Gateway status

Gateway logs

Live log stream

Gateway state events

Restart gateway

Stop gateway

Read gateway configuration

Update gateway configuration

Device pairing

List pending pairing requests

Approve a pairing request

Reject a pairing request

Revoke a paired device

List paired devices

Pairing event stream

Setup flow

Setup page

Save setup configuration

Get Ollama configuration

List Ollama models

Approve channel pairing via setup

Reset gateway

Export gateway data

Proxy error response (503)

WebSocket proxy

Terminal WebSocket messages

Environment variables

OpenClaw configuration and secrets

Default model configuration

Agent defaults

Tool configuration

Session configuration

Cron configuration

Logging configuration

Gateway settings

Health check

Persistent storage

Process management

Control UI URL resolution

URL construction

Example

WebSocket URL derivation

Per-agent gateway authentication

Token resolution order

Agent container configuration

OpenClaw configuration `env` section

Container environment variables

Gateway settings

Tool settings

Session settings

Agent defaults

Health monitoring

CORS

OpenAI-compatible endpoints