RLHF Feedback Loop | Hosted Guardrails and Shared Memory for AI Workflow Teams

The open-source RLHF Feedback Loop captures preference signals, generates prevention rules, and exports DPO-ready data for AI agents. Cloud Pro adds the hosted layer teams actually pay for: shared memory, provisioned API keys, funnel evidence, and team-safe workflow runs.

The best first paid wedge is not "agent infra" by itself. It is one workflow with a clear business outcome, such as lead-to-meeting, onboarding, or internal ops automation. This repo is the reliability layer behind that workflow.

Why someone would pay

• They want hosted API keys instead of self-hosting the feedback and guardrail store.
• They need shared memory and prevention rules across operators, repos, or agents.
• They need proof-ready runs and funnel evidence instead of local-only logs.

Quick Start

Add the MCP server directly in your client config:

Platform	Command
Claude	claude mcp add rlhf -- npx -y rlhf-feedback-loop serve
Codex	codex mcp add rlhf -- npx -y rlhf-feedback-loop serve
Gemini	gemini mcp add rlhf "npx -y rlhf-feedback-loop serve"
Amp	amp mcp add rlhf -- npx -y rlhf-feedback-loop serve
Cursor	cursor mcp add rlhf -- npx -y rlhf-feedback-loop serve

Optional auto-installer:

npx add-mcp rlhf-feedback-loop

The MCP is intentionally strict: a bare thumbs up or thumbs down is logged as a signal, but reusable memory promotion requires one sentence explaining why. If feedback is vague, the server asks for clarification instead of pretending it learned something.

OSS vs Cloud Pro

The OSS package stays free. Cloud Pro remains a low-friction founding offer while the hosted workflow layer proves onboarding and retention.

	OSS core	Cloud Pro
Price	$0	$10/mo
Feedback capture	Local MCP server	Hosted HTTPS API
Storage	Your machine	Managed cloud
DPO export	CLI command	API endpoint
Team sharing	Manual	Built-in
Onboarding	Self-serve	Checkout + provisioned API key

Landing Page | Get Cloud Pro ($10/mo) | Verification Evidence

Agent Runner Contract

This repo now ships a Symphony-compatible, repo-owned agent-runner contract:

• WORKFLOW.md: scope, proof-of-work, hard stops, and done criteria for isolated agent runs
• .github/ISSUE_TEMPLATE/ready-for-agent.yml: bounded intake template for "Ready for Agent" tickets
• .github/pull_request_template.md: proof-first handoff format for PRs

Validate the contract locally with:

node scripts/validate-workflow-contract.js
node scripts/prove-workflow-contract.js

Best First Use Case

The most credible first paid workflow is a lead-to-meeting system:

• inbound or CSV lead intake
• enrichment
• account research
• draft generation
• approval step
• CRM sync
• audit trail and prevention rules

Cloud Pro sits underneath that workflow as the hosted memory, guardrail, and evidence layer.

Architecture

Five-phase pipeline: Capture human signals → Validate with rubric engine → Learn via LanceDB vector memory → Prevent repeated mistakes → Export DPO pairs for fine-tuning.

Three-tier stack: external integrations (Claude, Codex, Gemini, ChatGPT via MCP/OpenAPI) → plugin orchestration (schema validation, Bayesian scoring, DPO export) → data persistence (JSONL, LanceDB vectors, ShieldCortex context packs).

Deep Dive

• GTM Revenue Wedge
• Pricing Research
• Verification Evidence
• API Reference
• Context Engine

License

MIT. See LICENSE.