Forge – Deterministic orchestrator for AI coding agents

Name: Forge – Deterministic orchestrator for AI coding agents
Availability: InStock
Author: lanath

by lanath·Feb 19, 2026·1 point·0 comments

Visit Project View on HN

AI Analysis

●●●BangerBig BrainSolve My ProblemShip It

Flips AI coding on its head: human gates decide, not the AI's confidence.

Strengths

•Inverts control flow elegantly—deterministic orchestration over AI autonomy is a real insight
•Multi-provider abstraction (Codex, Claude, Ollama) avoids vendor lock-in
•Self-hosted + PostgreSQL support means you own the plan history and audit trail

Weaknesses

•Extremely early (3-day sprint, 155 commits); untested at scale with complex repos
•Unproven whether deterministic gates catch AI hallucinations better than existing CI

Post Description

I built this in ~3 days because every AI coding tool lets the AI decide whether to run tests. Forge flips it: deterministic code orchestrates the AI, not the other way around. Your existing lint/test/coverage scripts become hard gates. AI writes code, your scripts decide if it ships. Provider-agnostic — ships with Codex SDK, Claude Agent SDK, Claude Code, and Ollama adapters. Forge built itself (the repo is the proof that the code produced is somewhat ok). Happy to answer questions about the architecture or get roasted on the code quality.