AIWorkHub

Local-first control plane for AI coding agents

Open Source

Developer Tools

GitHub

Visit WebsiteSee on Product HuntGithub ⧉Twitter ⧉

Hi Product Hunt 👋 I built AIWorkHub to solve problems I repeatedly encountered while using coding models on real, long-running software projects. The models could write code, but the engineering workflow around them kept breaking down: • important decisions disappeared after context compaction • agents repeatedly scanned the same repository • parallel workers could overlap or collide • task dependencies were difficult to coordinate • “finished” often meant a claim rather than verified evidence • switching between models meant losing continuity and operational control AIWorkHub turns every Git repository into an isolated, local-first AI engineering workspace. It provides a dependency-aware task DAG, structural Source Graph, durable project context, Session Manager, AI Memory, knowledge base, isolated model workers, callbacks, review inbox and evidence bundles containing diffs, tests, logs, artifacts and tool-use receipts. The manager and worker roles are deliberately separated. Workers cannot approve their own work. A task is accepted only after its evidence is independently reviewed. AIWorkHub works with models already available through VS Code or their authenticated CLIs, including Codex, Claude, Copilot, DeepSeek and GLM. It does not copy their credentials, require a separate AIWorkHub cloud account or move repository authority into a hosted service. This was not created as a multi-agent demo or a wrapper around several chat interfaces. I built it because I needed multiple coding models to work on serious repositories without losing decisions, conflicting with each other or asking me to trust unverifiable completion claims. AIWorkHub is open source under the MIT License and currently available through the VS Code Marketplace and GitHub Releases. I’d especially value feedback from developers who already use coding agents on large or long-running projects: What part of your current agent workflow loses the most time—context recovery, task coordination, model switching or review?