light-heart-labs--dreamserver/ods/extensions/services/openclaw

OpenClaw

⚠️ DEPRECATED as of 2026-05-12. Replaced by Hermes Agent (see extensions/services/hermes/ and docs/HERMES.md). New ODS installs no longer enable OpenClaw by default. This extension stays installable for one release cycle, then is removed. Migration notes: docs/MIGRATION-OPENCLAW-TO-HERMES.md.

Autonomous agent framework for ODS

Overview

OpenClaw is an autonomous AI agent gateway that orchestrates multi-step reasoning, tool use, and task execution using your local LLM. Agents can search the web via SearXNG, execute code, manage files, and chain complex workflows — all running locally without sending data to external services.

Features

• Autonomous agents: LLM-driven agents that plan, act, and iterate toward a goal
• Web search integration: Connects to local SearXNG for real-time web research
• Tool use: Agents can use a configurable set of tools across multiple steps
• Multi-model support: Primary inference model and a lighter bootstrap model for planning
• Gateway API: Exposes a REST API for launching and monitoring agent tasks
• LAN binding: Accessible from other devices on your local network

Dependencies

OpenClaw requires these services to be healthy before it starts:

Service	Role
searxng	Web search tool for agents
llama-server	LLM inference backend

Configuration

Environment variables (set in .env):

Variable	Default	Description
OPENCLAW_TOKEN	(required)	Gateway authentication token — must be set in .env
OPENCLAW_PORT	7860	External port for the OpenClaw web UI and API
LLM_MODEL	qwen3:30b-a3b	Primary LLM model for agent reasoning
BOOTSTRAP_MODEL	qwen3:8b-q4_K_M	Lighter model used for planning and bootstrapping
LLM_API_URL	http://llama-server:8080	Base URL of the LLM backend

OPENCLAW_TOKEN is required. The installer sets a random value in .env. If it is missing, the container will refuse to start. Verify it is set: grep OPENCLAW_TOKEN ods/.env

Volume Mounts

Host Path	Container Path	Purpose
./config/openclaw	/config (read-only)	Gateway config files and token injection script
./data/openclaw	/data	Agent task state and persistent data
./data/openclaw/home/agents	/home/node/.openclaw/agents	Agent definitions and sessions
./data/openclaw/home/canvas	/home/node/.openclaw/canvas	OpenClaw canvas state
./data/openclaw/home/cron	/home/node/.openclaw/cron	Scheduled agent jobs
./config/openclaw/workspace	/home/node/.openclaw/workspace	Agent workspace directory

Architecture

┌──────────┐  HTTP :7860     ┌──────────────┐
│ Browser  │────────────────▶│   OpenClaw   │
│  Client  │◀────────────────│   (Gateway)  │
└──────────┘  Agent results  └──────┬───────┘
                                    │ Plans tasks
                    ┌───────────────┴───────────────┐
                    ▼                               ▼
             ┌────────────┐                 ┌──────────────┐
             │  SearXNG   │                 │ llama-server │
             │ (Web Search│                 │    (LLM)     │
             └────────────┘                 └──────────────┘

Agent execution flow:

1. User submits a task via the web UI or API
2. OpenClaw uses the bootstrap model to decompose the task into steps
3. The primary LLM model executes each step, calling tools (search, code, etc.) as needed
4. Results are aggregated and returned to the user

Resource Limits

Limit	Value
CPU limit	2 cores
Memory limit	4 GB
CPU reservation	0.5 cores
Memory reservation	1 GB

Files

• manifest.yaml — Service metadata (port, health endpoint, required env vars)
• compose.yaml — Container definition (image, environment, volumes, resource limits)

Troubleshooting

OpenClaw refuses to start (OPENCLAW_TOKEN error):

# Check token is set
grep OPENCLAW_TOKEN ods/.env

# Set a token value if missing
echo "OPENCLAW_TOKEN=$(openssl rand -hex 24)" >> ods/.env
docker compose up -d openclaw

OpenClaw not starting (SearXNG dependency):

OpenClaw waits for SearXNG to be healthy. Check SearXNG first:

docker compose ps ods-searxng
docker compose logs ods-searxng

Cannot access web UI on port 7860:

docker compose ps ods-openclaw
docker compose logs ods-openclaw

Agents not using the right model:

• Update LLM_MODEL in .env to match a model loaded in llama-server
• Restart OpenClaw: docker compose restart openclaw

Token injection errors:

# Check config directory
ls ods/config/openclaw/
docker compose logs ods-openclaw | grep -i token