m0n0x41d/haft — Engineering decisions engine that know when they're stale.

Description and use

**Brief Summary**
m0n0x41d/haft is an open‑source “engineering‑decisions engine” that tracks the freshness of AI‑generated evidence, enforces parity across model outputs, and lets you frame, compare, and decide on model responses. It ships ready‑to‑use integrations for Claude, Cursor, Gemini CLI, Codex and other LLMs, making it easy to bolt AI‑driven decision logic into existing tools.

**Value**

• **Accelerates AI feature development** – you get a reusable decision layer (evidence decay, parity checks, RAG/agent orchestration) instead of building one from scratch.
• **Improves reliability** – automatic staleness detection and parity enforcement reduce hallucinations and inconsistent outputs across multiple model providers.
• **Vendor‑agnostic** – the same engine works with several major LLM APIs, so you can experiment or switch providers without rewriting logic.

**Practical Adoption Path**

1. **Clone the repo** (Go 1.22+ required) and run the provided Dockerfile or binary to start the Haft service locally.
2. **Configure connectors** via the YAML/JSON manifest – specify API keys, target models (Claude, Gemini, Codex, etc.), and any custom evidence sources (databases, vector stores).
3. **Instrument your code** – import the Go SDK or call the REST/CLI endpoint from your existing application, passing the “frame” (question), “compare” (candidate responses), and “decide” (policy) payloads.
4. **Iterate** – use the built‑in monitoring UI or logs to see evidence decay scores and parity violations, then tune policies or add new data sources.
5. **Promote to production** – containerize the service, place it behind a service mesh or API gateway, and enable TLS/role‑based access control using the supplied Helm chart.

**Production Readiness**

• **Activity & community** – 1,322 ★, 104 forks, recent commits (last update 2026‑05‑13), and a clear Go codebase indicate an active project.
• **Maturity** – the core engine, CLI, and SDK are stable; multiple LLM integrations have been tested, and the repository includes CI pipelines and Helm charts for deployment.
• **Risk profile** – no major licensing or security red flags identified, though a final audit of the license (MIT‑style) and supply‑chain dependencies is advisable.
• **Fit for pilot** – given its high signal‑to‑noise ratio, strong adoption signals, and straightforward integration steps, Haft is a solid candidate for a production pilot in environments that need trustworthy, multi‑model AI decision making.

Problem

m0n0x41d/haft helps add AI capability without starting from a blank model stack.

Outcome

High for an OSS candidate: recent activity, adoption, and ecosystem signals are strong enough for a serious pilot.

Workflow

• prototype AI features
• build RAG or agent workflows
• evaluate model tooling

Evidence

Engineering decisions engine that know when they're stale. Frame, compare, decide — with evidence decay and parity enforcement. For Claude Code, Cursor, Gemini CLI, Codex and more.

Source