All 61 recipes
Browser LLM Stack
Browser-native model work without pretending every useful workflow has a clean API.
Claude Code via tmux Relay
How to let OpenClaw drive Claude Code through an interactive tmux session, both for second-opinion review and for scripted one-shot calls, keeping Claude in its first-party harness rather than…
Updated 2026-06-29Compaction & Context Tuning
How to configure OpenClaw's compaction, memory flush, context pruning, and session search so your agent doesn't lose its mind (or personality) during long sessions.
Updated 2026-06-05Local LLM Fallback
Use local models for boring, bounded work so your paid models stay available for judgment.
Updated 2026-05-11Migrating from claude-cli to ACP
How to move Claude Opus off the main-agent slot and onto an ACP escalation path after Anthropic blocked subscription OAuth from third-party harnesses in April 2026.
Updated 2026-06-29Model Retirement and Fallbacks
When a model id is retired or an alias changes, a long-lived agent stack with dozens of cron jobs and several agents can break silently. This is the playbook for finding every place a model id hides,…
Updated 2026-06-10Multi-Model Orchestration
How to run multiple AI models in one OpenClaw setup, assign each to the right task tier, and stop burning expensive tokens on work that doesn't need them.
Updated 2026-06-05OAuth and Token Lifecycle for a Multi-Provider Agent Stack
A long-lived agent stack runs on subscription OAuth, not API keys, and the recurring pain is not auth setup. It is token lifecycle: rotation, caching, fallback, and the silent failures that look like…
Updated 2026-06-10Orchestrating with GPT 5.5: Narration Guards and Strict-Agentic Gaps
Running GPT 5.5 (via OpenAI Codex Pro) as your main orchestrator is cheap and fast compared to frontier API billing, but it has specific failure modes that will quietly eat hours of your time until…
Updated 2026-06-10Prompt Caching: Maximize Cache Hits, Minimize Token Costs
How OpenClaw's prompt caching works across providers, how to keep your cache hit rate high, and the anti-patterns that silently cost you money every turn.
Updated 2026-06-10Running Claude Code in OpenClaw via ACP
In April 2026, Anthropic blocked subscription OAuth from third-party harnesses. The anthropic:claude-cli backend that most OpenClaw users had plugged a Max subscription into stopped working…
Updated 2026-06-30Self-Improving Agents
How to build an AI agent that learns from corrections, captures mistakes as institutional knowledge, runs automated memory sweeps, and gets better over time instead of repeating the same errors.
Updated 2026-06-10Session Management: Why Your Chat App Is Holding You Back
How to manage OpenClaw sessions effectively using Discord (or similar channel-based platforms) instead of single-thread messaging apps. The difference between productive multi-project orchestration…
Updated 2026-06-10Skills Development
How to write custom OpenClaw skills, structure them for discoverability, and extend your agent's capabilities with reusable task-specific instructions.
Updated 2026-06-10Sub-Agent Patterns: Orchestration, Spawning, and Gotchas
How to use OpenClaw sub-agents effectively. Spawn patterns, model assignment, error handling, and the lessons we learned from breaking things.
Updated 2026-06-29Cron Job Patterns
How to schedule automated tasks in OpenClaw, assign the right model to each job, batch checks into heartbeats, and avoid the pitfalls that waste tokens and break silently.
Updated 2026-06-10Cron Patterns
The three-layer cron stack I actually run: systemd timers for OS plumbing, OpenClaw cron for single-shot agent tasks, n8n schedule triggers for multi-step workflows. Pick the wrong layer and you'll…
Hooks
Hooks enforce policy at boundaries and shape behavior at seams. Pick the wrong layer and you'll scrub private DMs, fight your orchestrator's hook contract, or watch async work disappear into a…
Multi-Channel Setup
How OpenClaw handles multiple messaging platforms simultaneously, session isolation between channels, and practical patterns for Discord, Telegram, and Signal.
Updated 2026-06-10n8n Failure Classifier
One Error Trigger workflow, wired as errorWorkflow on everything active, that turns raw stack traces into buckets and routes each bucket to the action it actually deserves. Without it a…
Updated 2026-06-04n8n Patterns
The interface surfaces, sandbox traps, and failure-routing patterns that actually matter for running n8n as the multi-step workflow layer underneath an agent stack. Pick the wrong API surface and…
Sandbox Shims
Put cheap wrappers in front of risky commands so worker lanes fail closed before they touch the network, git remote, or host secrets.
Updated 2026-05-11Self-Hosted Social Publishing Stack
The publishing plumbing I actually run: Postiz for scheduling and fan-out to the networks, n8n for the multi-step orchestration around it, and an agent driving both over MCP with writes gated off by…
Updated 2026-06-10AdGuard Home as a Network DNS Sinkhole
Block ads, trackers, and malware at the resolver instead of on every device, run a synced standby so a single reboot doesn't take DNS down, and let an AI agent query and tune it through tiered MCP…
Updated 2026-06-30Backup & Recovery
How to protect your OpenClaw workspace, configuration, and memory from data loss. Encrypted backups, restore testing, and disaster recovery planning.
Updated 2026-05-31Desktop Integration: The Daily Driver as a Peer
Most homelab writeups treat the desktop as a dumb client that connects to the server. I run it the other way too. My always-on Linux agent host SSHes into the Windows 11 daily driver, mounts its…
Updated 2026-06-04Homelab Topology: The Map
This is the floor plan of my homelab. One hypervisor, a pile of LXC containers, a couple of VMs, and a backup server that lives on the same box it protects. It covers what runs where, how I split LXC…
Updated 2026-06-04NAS and Network Storage Mounts
How to wire network storage into a Linux host so a powered-off peer never hangs your boot, an agent never deletes irreplaceable data, and a consumer NAS never silently corrupts a backup chunkstore.…
Updated 2026-06-04OpenClaw Host Topology
A production agent host is not just one daemon. It is config, channels, cron, memory, browser automation, plugins, health checks, and the boring glue that keeps all of it observable.
Proxmox as the Agent Lab Substrate
One consumer-grade hypervisor runs the always-on agent stack's supporting services, an AI agent inspects and safely operates it through a token-gated MCP, and a separate auditor grades the host…
Updated 2026-06-10Service Isolation: One Service Per Container
I run my homelab as a pile of small, boring, unprivileged LXC containers. One service each. Not one fat VM with a docker-compose monolith, not a single Debian box hand-fed twelve daemons. This guide…
Updated 2026-06-04Upgrade Hygiene: Surviving openclaw update
Every OpenClaw minor release has, at least once, silently regenerated my systemd unit and dropped custom directives. If you don't plan for this, the gateway crash-loops at 4am and you find out over…
Updated 2026-05-06Agent Incident Runbook
When an agent breaks something, stop the bleeding first, preserve evidence second, and only then fix the root cause.
Updated 2026-05-11Agent Security Hardening
How to treat your AI agent as an untrusted actor and build guardrails that actually work. Includes a real post-mortem from when a sub-agent nuked a production database.
Updated 2026-06-10MCP-Driven Incident Response: One Operator, One Agent, A Whole SOC
A self-hosted SOC where an AI agent drives incident response through MCP servers you wrote yourself. A SIEM alert fires, the agent triages it, opens a case, runs observable analysis, enriches…
Updated 2026-06-10Secret Management
Secrets belong in narrow local stores with boring permissions, not in prompts, repos, screenshots, or memory.
Updated 2026-07-13Security Hardening: Linux Host for OpenClaw
Practical hardening runbook for an Ubuntu 24.04 machine running OpenClaw as an always-on AI agent. This covers firewall configuration, SSH lockdown, fail2ban, and service binding to reduce attack…
Updated 2026-04-19Security Hardening: Windows + WSL2 Host for OpenClaw
Practical hardening runbook for a Windows machine running OpenClaw inside WSL2. Covers Windows Firewall, RDP/SSH/SMB lockdown, port proxy hygiene, WSL-specific gotchas, and defense-in-depth for a…
Updated 2026-04-19Wazuh Triage: RCA, Fix, Narrow Suppress, One Pass
When a Wazuh alert fires, find the root cause first, fix the underlying problem, and only then write the narrowest possible suppression. Do all three in one pass so the alert channel stays…
Updated 2026-06-04Bootstrap Files: What Each Agent File Owns
Your agent's personality, safety rails, memory index, and local runbook should not all live in one giant prompt. Split them by job so compaction, cache invalidation, and handoffs stay sane.
Claude Code and Codex Memory Handoffs: A Sync Path Into OpenClaw
If you run Claude Code or Codex locally alongside an OpenClaw gateway, you end up with multiple session memories. This guide describes the shared handoff format and ingester that keep OpenClaw as the…
Updated 2026-07-09Memory Architecture
Memory in an agent stack is a layered store of point-in-time claims, not live state. Pick the wrong relationship between memory and current reality and your agent will confidently recommend file…
Memory Management & Token Optimization
How to build an AI agent memory system that doesn't eat your context window alive. From raw conversation history to semantic search with local embeddings.
Updated 2026-06-10Obsidian Sync Without Conflict Roulette
One vault, one sync layer, one place for automation to write. That is how you get bidirectional cloud sync without waking up to three "conflicted copy" files and a broken notes index.
Session JSONL as Memory Source, Not Noise
Transcript logs are an audit trail and a mining seam. Treat them as search-only source material, not something to shovel raw into every prompt.
The MiseLedger Evidence Pipeline: Searchable Work History From Session Logs
Agent sessions are useful until you need to find a decision from six weeks ago. MiseLedger imports local harness logs into one SQLite archive, indexes them with FTS5, and returns citable evidence…
Updated 2026-07-13Bare-Metal Agent Host
The primary Linux box in the fleet, picked once, lives for years. Spec it for the workload that already gives you grief, not the workload you imagine you'll have someday.
Disk Layout with LVM
Two NVMe drives, one volume group, one growable home. The installer's default LVM is a trap. Lay this out by hand once and you'll never have to resize on a Sunday afternoon.
Kernel Tuning for an Always-On AI Host
Six sysctl knobs, one I/O scheduler decision, and a swap policy. Almost everything else the kernel ships with is fine. The defaults are tuned for "every workload," which means they are tuned for none.
Brigade
Install and operate the cookbook's agent workspace shape instead of copying it by hand: bootstrap files, per-writer memory handoffs, content guards, MCP sync, local work loops, operator checks,…
MCP Catalog
Every MCP server published from this stack, what each one wraps, and where it fits. Most exist because a service the agent needed to talk to did not have a maintained MCP yet, so I wrote one.
MCP READMEs: All Five Clients
Every MCP repo in the catalog ships with the same five-block README. If a setup block is missing, half your potential users walk away. Five clients, one shape, copy-paste tested against each.
OpsDeck
A self-hosted dashboard that surfaces the parts of the stack you'd otherwise check by opening five terminals. Twenty routes, a companion API on port 8005, and one place to inspect projects, services,…
Repo Redeploy
One cron job, every ten minutes, watches your own MCP servers and CLI tools for new commits on main and quietly redeploys them in place. Push from anywhere, the change is live within ten minutes.
Skillet
The skills rack: installable agent skills for auditing, improving, and shipping repos. The flagship is line-check, a repo audit that ends in a leverage-sorted improvement backlog instead of a wall of…
Manifesto vs Framework
A cookbook is not a tool. It is a record. The difference matters because tools are owed support and frameworks are owed compatibility; cookbooks are owed only that the recipes still work for the…
What This Stack Is Not
Easier to write a cookbook by listing what is not in it. Every entry below is a thing other people put in their stacks, that I tried, considered, or watched fail, and explicitly do not do here.
Why Dogfood Everything
Ship it. Use it. Break it. Fix it. Write it down. The loop is short on purpose. Anything in this cookbook that did not survive that loop is not in this cookbook.
Why One Control Host
A bare-metal fleet is fine. A split-brain agent is not. The trade is not "single box forever" versus "real infrastructure." The trade is one canonical control point for memory, receipts, cron, and…
No recipes match that filter.