Dhee is an open-source context firewall for AI coding agents. It decides what agents should see, remember, forget, compress, and expand each turn, so Claude Code, Codex, Cursor and other MCP clients stop drowning in stale files, logs and transcripts. It runs locally, supports repo-shared context, handoff state, evidence pointers, Dhee UI, and reports 99.4% R@5 on LongMemEval.
Hey Product Hunt 👋
I built Dhee because AI coding agents are getting powerful, but their context is still messy.
Claude Code, Codex, Cursor and other agents can write code fast, but they keep rereading the same files, lose decisions after compaction, carry stale context, and flood the model with huge logs or diffs.
Dhee is my attempt to fix that. It runs locally beside your coding agent and acts like a context firewall: deciding what the agent should see, remember, forget, compress, or expand each turn.
It supports MCP, Claude Code, Codex, repo-shared context, handoff state, evidence pointers, and a local UI to inspect what was injected or hidden.
The goal is simple: your agent should not see everything. It should see the right thing, with proof.
Would love feedback from builders using AI coding agents daily.
Report
No reviews yetBe the first to leave a review for Dhee