Building CLI backend plugins

CLI backend plugins let OpenClaw call a local AI CLI as a text inference backend. The backend appears as a provider prefix in model refs:

acme-cli/acme-large

Use a CLI backend when the upstream integration is already exposed as a local command, when the CLI owns local login state, or as a fallback when API providers are unavailable.

If the upstream service exposes a normal HTTP model API, write a provider plugin instead. If the upstream runtime owns complete agent sessions, tool events, compaction, or background task state, use an agent harness.

What the plugin owns

A CLI backend plugin has three contracts:

Contract	File	Purpose
Package entry	`package.json`	Points OpenClaw at the plugin runtime module
Manifest ownership	`openclaw.plugin.json`	Declares the backend id before runtime loads
Runtime registration	`index.ts`	Calls `api.registerCliBackend(...)` with command defaults

The manifest is discovery metadata: it does not execute the CLI or register runtime behavior. Runtime behavior starts when the plugin entry calls api.registerCliBackend(...).

Minimal backend plugin

Create package metadata

package.json

{
  "name": "@acme/openclaw-acme-cli",
  "version": "1.0.0",
  "type": "module",
  "openclaw": {
    "extensions": ["./index.ts"],
    "compat": {
      "pluginApi": ">=2026.3.24-beta.2",
      "minGatewayVersion": "2026.3.24-beta.2"
    },
    "build": {
      "openclawVersion": "2026.3.24-beta.2",
      "pluginSdkVersion": "2026.3.24-beta.2"
    }
  },
  "dependencies": {
    "openclaw": "^2026.3.24"
  },
  "devDependencies": {
    "typescript": "^5.9.0"
  }
}

Published packages must ship built JavaScript runtime files. If your source entry is ./src/index.ts, add openclaw.runtimeExtensions pointing at the built JavaScript peer. See Entry points.

Declare backend ownership

openclaw.plugin.json

{
  "id": "acme-cli",
  "name": "Acme CLI",
  "description": "Run Acme's local AI CLI through OpenClaw",
  "cliBackends": ["acme-cli"],
  "setup": {
    "cliBackends": ["acme-cli"],
    "requiresRuntime": false
  },
  "activation": {
    "onStartup": false
  },
  "configSchema": {
    "type": "object",
    "additionalProperties": false
  }
}

cliBackends is the runtime ownership list; it lets OpenClaw auto-load the plugin when config or model selection mentions acme-cli/....setup.cliBackends is the descriptor-first setup surface. Add it when model discovery, onboarding, or status should recognize the backend without loading plugin runtime. Use requiresRuntime: false only when those static descriptors are enough for setup.

index.ts

import { definePluginEntry } from "openclaw/plugin-sdk/plugin-entry";
import {
  CLI_FRESH_WATCHDOG_DEFAULTS,
  CLI_RESUME_WATCHDOG_DEFAULTS,
  type CliBackendPlugin,
} from "openclaw/plugin-sdk/cli-backend";

function buildAcmeCliBackend(): CliBackendPlugin {
  return {
    id: "acme-cli",
    liveTest: {
      defaultModelRef: "acme-cli/acme-large",
      defaultImageProbe: false,
      defaultMcpProbe: false,
      docker: {
        npmPackage: "@acme/acme-cli",
        binaryName: "acme",
      },
    },
    config: {
      command: "acme",
      args: ["chat", "--json"],
      output: "json",
      input: "stdin",
      modelArg: "--model",
      sessionArg: "--session",
      sessionMode: "existing",
      sessionIdFields: ["session_id", "conversation_id"],
      systemPromptFileArg: "--system-file",
      systemPromptWhen: "first",
      imageArg: "--image",
      imageMode: "repeat",
      reliability: {
        watchdog: {
          fresh: { ...CLI_FRESH_WATCHDOG_DEFAULTS },
          resume: { ...CLI_RESUME_WATCHDOG_DEFAULTS },
        },
      },
      serialize: true,
    },
  };
}

export default definePluginEntry({
  id: "acme-cli",
  name: "Acme CLI",
  description: "Run Acme's local AI CLI through OpenClaw",
  register(api) {
    api.registerCliBackend(buildAcmeCliBackend());
  },
});

The backend id must match the manifest cliBackends entry. The registered config is only the default; user config under agents.defaults.cliBackends.acme-cli merges over it at runtime.

Config shape

CliBackendConfig describes how OpenClaw should launch and parse the CLI:

Field	Use
`command`	Binary name or absolute command path
`args`	Base argv for fresh runs
`resumeArgs`	Alternate argv for resumed sessions; supports `{sessionId}`
`output` / `resumeOutput`	Parser: `json`, `jsonl`, or `text`
`jsonlDialect`	JSONL event dialect: `claude-stream-json` or `gemini-stream-json`
`liveSession`	Long-lived CLI process mode (`claude-stdio`)
`input`	Prompt transport: `arg` or `stdin`
`maxPromptArgChars`	Max prompt length for `arg` mode before falling back to stdin
`env` / `clearEnv`	Extra env vars to inject, or names to strip before launch
`modelArg`	Flag used before the model id
`modelAliases`	Map OpenClaw model ids to CLI-native ids
`sessionArg` / `sessionArgs`	How to pass a session id
`sessionMode`	`always`, `existing`, or `none`
`sessionIdFields`	JSON fields OpenClaw reads from CLI output
`systemPromptArg` / `systemPromptFileArg`	System prompt transport
`systemPromptFileConfigArg` / `systemPromptFileConfigKey`	Config-override transport for a system prompt file (for example `-c`)
`systemPromptMode`	`append` or `replace`
`systemPromptWhen`	`first`, `always`, or `never`
`imageArg` / `imageMode`	Image path flag and how to pass multiple images (`repeat` or `list`)
`imagePathScope`	Where staged image files live before handoff: `temp` or `workspace`
`serialize`	Keep same-backend runs ordered
`reseedFromRawTranscriptWhenUncompacted`	Opt in to bounded raw-transcript reseed before compaction for safe session resets
`reliability.outputLimits`	Max raw JSONL chars/lines retained for one live CLI turn (live-session backends)
`reliability.watchdog`	No-output timeout tuning, separate for fresh vs resumed runs

Prefer the smallest static config that matches the CLI. Add plugin callbacks only for behavior that really belongs to the backend.

Advanced backend hooks

CliBackendPlugin can also define:

Hook	Use
`normalizeConfig(config, context)`	Rewrite legacy user config after merge
`resolveExecutionArgs(ctx)`	Add request-scoped flags such as thinking effort or side-question isolation
`prepareExecution(ctx)`	Create temporary auth or config bridges before launch
`transformSystemPrompt(ctx)`	Apply a final CLI-specific system prompt transform
`textTransforms`	Bidirectional prompt/output replacements
`defaultAuthProfileId`	Prefer a specific OpenClaw auth profile
`authEpochMode`	Decide how auth changes invalidate stored CLI sessions
`nativeToolMode`	Declare whether the CLI has always-on native tools
`sideQuestionToolMode`	Declare disabled native tools for `/btw` side questions
`bundleMcp` / `bundleMcpMode`	Opt into OpenClaw’s loopback MCP tool bridge
`ownsNativeCompaction`	Backend owns its own compaction - OpenClaw defers

Keep these hooks provider-owned. Do not add CLI-specific branches to core when a backend hook can express the behavior. ctx.executionMode is "agent" for normal turns and "side-question" for ephemeral /btw calls. Use it when the CLI needs different one-shot flags, such as disabling native tools, session persistence, or resume behavior for BTW. If a backend normally has nativeToolMode: "always-on" but its side-question argv reliably disables those tools, also set sideQuestionToolMode: "disabled"; otherwise OpenClaw fails closed when BTW requires a no-tools CLI run.

`ownsNativeCompaction`: opting out of OpenClaw compaction

If your backend runs an agent that compacts its own transcript, set ownsNativeCompaction: true so OpenClaw’s safeguard summarizer never runs against its sessions - the CLI compaction lifecycle returns a no-op and the turn proceeds. claude-cli declares it because Claude Code compacts internally with no harness endpoint. Native-harness sessions such as Codex keep routing to their harness compaction endpoint instead. Only declare it when all of the following hold, or a deferred over-budget session can stay over budget or go stale (OpenClaw no longer rescues it):

the backend reliably compacts or bounds its own transcript as it nears its window;
it persists a resumable session so the compacted state survives turns (for example --resume / --session-id);
it is not a native-harness compaction session - matching agentHarnessId sessions route to the harness endpoint instead.

MCP tool bridge

CLI backends do not receive OpenClaw tools by default. If the CLI can consume an MCP configuration, opt in explicitly:

return {
  id: "acme-cli",
  bundleMcp: true,
  bundleMcpMode: "codex-config-overrides",
  config: {
    command: "acme",
    args: ["chat", "--json"],
    output: "json",
  },
};

Supported bridge modes:

Mode	Use
`claude-config-file`	CLIs that accept an MCP config file
`codex-config-overrides`	CLIs that accept config overrides on argv
`gemini-system-settings`	CLIs that read MCP settings from their system settings directory

Only enable the bridge when the CLI can actually consume it. If the CLI has its own built-in tool layer that cannot be disabled, set nativeToolMode: "always-on" so OpenClaw can fail closed when a caller requires no native tools.

User configuration

Users can override any backend default:

{
  agents: {
    defaults: {
      cliBackends: {
        "acme-cli": {
          command: "/opt/acme/bin/acme",
          args: ["chat", "--json", "--profile", "work"],
          modelAliases: {
            large: "acme-large-2026",
          },
        },
      },
      model: {
        primary: "openai/gpt-5.5",
        fallbacks: ["acme-cli/large"],
      },
    },
  },
}

Document the minimum override users are likely to need - usually only command when the binary is outside PATH.

Verification

For bundled plugins, add a focused test around the builder and setup registration, then run the plugin’s targeted test lane:

pnpm test extensions/acme-cli

For local or installed plugins, verify discovery and one real model run:

openclaw plugins inspect acme-cli --runtime --json
openclaw agent --message "reply exactly: backend ok" --model acme-cli/acme-large

If the backend supports images or MCP, add a live smoke that proves those paths with the real CLI. Do not rely on static inspection for prompt, image, MCP, or session-resume behavior.

Checklist

package.json has openclaw.extensions and built runtime entries for published packages

openclaw.plugin.json declares cliBackends and intentional activation.onStartup

setup.cliBackends is present when setup/model discovery should see the backend cold

api.registerCliBackend(...) uses the same backend id as the manifest

User overrides under agents.defaults.cliBackends.<id> still win

Session, system prompt, image, and output parser settings match the real CLI contract

Targeted tests and at least one live CLI smoke prove the backend path

CLI backends - user configuration and runtime behavior
Building plugins - package and manifest basics
Plugin SDK overview - registration API reference
Plugin manifest - cliBackends and setup descriptors
Agent harness - full external agent runtimes

​What the plugin owns

​Minimal backend plugin

​Config shape

​Advanced backend hooks

​ownsNativeCompaction: opting out of OpenClaw compaction

​MCP tool bridge

​User configuration

​Verification

​Checklist

​Related

What the plugin owns

Minimal backend plugin

Config shape

Advanced backend hooks

`ownsNativeCompaction`: opting out of OpenClaw compaction

MCP tool bridge

User configuration

Verification

Checklist

Related