ProdClaw

ProdClaw turns an existing OpenClaw setup into a reliable, cost-efficient personal AI coworker.

Your AI agent should not forget, drift, or silently break. ProdClaw adds an opinionated production harness around OpenClaw: durable memory through LanceDB Pro, OpenRouter model routing, main/consultant/compliance role separation, compliance Slack delivery, scheduled self-audits, safe backup, and managed-file rollback.

ProdClaw v1 is for existing OpenClaw users who want a production-ready setup. It is not a from-scratch OpenClaw installer yet.

ProdClaw does not replace your whole OpenClaw home or take ownership of runtime data. It adds a managed production harness around the parts it owns, while preserving conversations, memory, logs, auth, runtime databases, non-ProdClaw skills, and user-created files.

Who v1 is for

ProdClaw v1 is for existing OpenClaw users, technical founders, indie hackers, AI power users, and small technical teams who want a stricter production harness around their local agent runtime.

It assumes you are comfortable with OpenRouter, Slack apps, and an existing OpenClaw installation.

If you do not already have OpenClaw installed, wait for the v2 from-scratch setup path.

Requirements

ProdClaw v1 requires:

• an existing OpenClaw installation;
• OpenRouter for model routing;
• LanceDB Pro for durable semantic memory;
• a compliance Slack app for audit delivery.

A main Slack app is optional. You can skip it if you prefer to interact through your existing OpenClaw dashboard/local UI.

If LanceDB Pro is already installed and healthy, ProdClaw keeps it and preserves existing data. If it is missing, disabled, or not bound as the memory slot, ProdClaw should stop and ask for an approved install/enable/bind path before apply or cron enablement.

ProdClaw does not require a separate database server. OpenClaw runtime storage and LanceDB Pro memory are local/embedded storage concerns.

What ProdClaw adds

Layer	v1 decision
Runtime	Existing OpenClaw installation, migrated in place after backup
Agents	main, consultant, and compliance roles
Models	OpenRouter-first routing with curated primary/fallback policies
Memory	LanceDB Pro as the durable semantic memory layer
Slack	Required compliance Slack delivery; optional main Slack interaction
Cron	Scheduled compliance/self-audit jobs after setup safety checks
Safety	Managed-file diff, backup, and rollback boundaries

Memory without agent overhead

ProdClaw treats memory as infrastructure, not as another chore for the agent.

The agent should not have to manually decide what to remember, where to store it, how to search it, or how to inject it back into context. LanceDB Pro makes memory part of the system layer: auto-capture, hybrid retrieval, reranking, multi-scope isolation, long-context chunking, and recall happen behind the harness so the agent can focus on the task instead of managing its own notebook.

That is the product bet: memory continuity should improve reliability without bloating every prompt or forcing the agent to do extra memory work. LanceDB Pro is the memory engine behind that bet, with LoCoMo-style long-context memory benchmarking positioned around an 80% score.

Cost-efficient by design

ProdClaw is production-ready without being wasteful. Routine work can run on efficient models, while high-judgment work escalates through the consultant architecture. Compliance jobs use curated model/fallback policies so you are not using the most expensive model for every task.

Why this harness exists

Most agent setups optimize the wrong thing. They add more prompt, more tools, more providers, more autonomy, and then wonder why the system gets slower and less predictable.

ProdClaw takes the opposite position:

• the harness should be thin;
• the skills should be fat;
• model routing should be centralized;
• memory should be a real subsystem;
• recurring audits should be scheduled, not remembered;
• Slack should be a controlled delivery path, not an unbounded permission surface.

The default is intentionally opinionated because the value is reliability, memory continuity, compliance checks, safer upgrades, rollback, and cost-efficient model routing.

Before / After

Without this harness, OpenClaw can grow by accident: provider plugins stay enabled, prompts accrete, cron jobs drift, memory becomes ambiguous, and every new feature adds more context weight.

With ProdClaw, the runtime has a job:

• main handles owner-facing orchestration.
• consultant handles high-judgment architecture and review.
• compliance audits the harness on a schedule.
• LanceDB Pro owns durable semantic memory.
• OpenRouter owns model routing.
• Compliance Slack owns production audit delivery.

The design goal is simple: keep the harness thin, keep skills fat, and keep recurring operational discipline outside the user's head.

Install path

Primary path: existing OpenClaw users.

Recommended flow:

npm run prodclaw -- inspect --home ~/.openclaw
npm run prodclaw -- configure --home ~/.openclaw --interactive --config-out local/prodclaw.configure.json
npm run prodclaw -- render --home ~/.openclaw --out ./rendered --config local/prodclaw.configure.json
npm run prodclaw -- validate --rendered ./rendered
npm run prodclaw -- doctor --home ~/.openclaw --rendered ./rendered
npm run prodclaw -- diff --home ~/.openclaw --rendered ./rendered
npm run prodclaw -- registration-plan --home ~/.openclaw --rendered ./rendered
npm run prodclaw -- apply --home ~/.openclaw --rendered ./rendered --confirm APPLY
npm run prodclaw -- register-runtime --home ~/.openclaw --rendered ./rendered
npm run prodclaw -- smoke-slack --home ~/.openclaw --delivery-confirmed
npm run prodclaw -- enable-cron --home ~/.openclaw --rendered ./rendered --yes

Important safety gates:

• apply requires typed --confirm APPLY unless using legacy non-interactive --yes.
• register-runtime defaults to dry-run and executes only with --confirm REGISTER.
• smoke-slack writes a smoke result only after explicit operator delivery confirmation.
• enable-cron requires doctor plus a passed compliance Slack smoke result.
• Cron jobs are registered/kept disabled until enable-cron.
• No command runs cron during setup.
• No command restarts the gateway automatically.

The v1 command surface is:

prodclaw inspect
prodclaw configure
prodclaw render
prodclaw validate
prodclaw doctor
prodclaw diff
prodclaw registration-plan
prodclaw apply
prodclaw register-runtime
prodclaw smoke-slack
prodclaw enable-cron
prodclaw revert
prodclaw scan-secrets

From-scratch OpenClaw setup is intentionally deferred. See docs/FROM_SCRATCH_DEFERRED.md.

Runtime registration boundary

ProdClaw writes managed static files through the backup/diff/apply path. OpenClaw-owned runtime registration should use OpenClaw CLI where safe.

Current status:

• prodclaw registration-plan shows intended runtime actions.
• prodclaw register-runtime uses confirmed OpenClaw CLI syntax to add missing ProdClaw-owned compliance cron jobs.
• Cron jobs are added disabled.
• Missing-agent creation is blocked unless OpenClaw exposes or confirms a deterministic stable-ID path.
• The deterministic agent-ID follow-up is tracked separately.

User profile rendering

ProdClaw keeps the harness opinionated. Initial setup only personalizes the user's name and timezone. The operating style, role framing, and rules remain part of the harness.

Validation

For staged output checks, run:

npm run prodclaw -- validate --rendered ./rendered

For repository hygiene checks, run:

npm run prodclaw -- scan-secrets

Agent-led setup

If an agent is doing the install, start with INSTALL_FOR_AGENTS.md. The agent must inspect first, configure/render next, validate and doctor before apply, show diff and registration-plan before write actions, apply only after explicit approval, and enable cron only after compliance Slack delivery is proven.

Design documents

• Reference Architecture
• Existing OpenClaw Migration
• What ProdClaw Touches
• Security And Slack
• Operations
• Runtime Registration
• Secret Scanning
• Smoke Results
• Apply and Revert Confirmation
• From-Scratch Setup Deferred

Safety boundary

ProdClaw does not replace the full OpenClaw home. It manages the harness layer only:

• OpenClaw config;
• workspace instructions;
• consultant workspace instructions;
• compliance workspace instructions;
• compliance prompts;
• ProdClaw-owned skills;
• model routing policy;
• plugin policy;
• ProdClaw-owned cron jobs.

ProdClaw does not modify or delete:

• sessions;
• conversations;
• memories;
• LanceDB data;
• markdown memory;
• logs;
• auth profiles;
• runtime databases;
• plugin installation directories;
• non-ProdClaw skills;
• non-ProdClaw cron jobs;
• user-created files outside managed paths.

Before applying changes, ProdClaw creates a backup and shows a diff. Nothing is applied until you explicitly approve it.

Revert

Preview first:

npm run prodclaw -- revert --home ~/.openclaw --dry-run

Then, after review:

npm run prodclaw -- revert --home ~/.openclaw --confirm REVERT

prodclaw revert restores only ProdClaw-managed harness files from managed backups. It does not roll back conversations, sessions, memories, LanceDB data, logs, auth profiles, runtime databases, non-ProdClaw skills, non-ProdClaw cron jobs, or user-created files.

Full OpenClaw home restore is intentionally not part of v1.