Question 1

What is Dhee?

Accepted Answer

Dhee is an open-source, self-evolving memory and context-router for LLM-powered AI agents. It sits between your agent (Claude Code, Cursor, Codex, Gemini CLI, Aider, Cline, Goose, or any MCP client) and the model, chunks and indexes your CLAUDE.md/AGENTS.md/skills library into decay-aware memory, digests fat tool output at the source, and tunes its own retrieval policy over time.

Question 2

How does Dhee reduce Claude Code token usage?

Accepted Answer

Two mechanisms. First, a 500-line CLAUDE.md that normally re-loads every turn becomes a vector-indexed memory that injects only matching rules — about 240 tokens instead of 5,700. Second, the router wraps Read and Bash so a 10 MB git log becomes a 40-token digest with a pointer; raw content only re-enters context if the model explicitly expands the pointer. Typical reduction is 90%+.

Question 3

How is Dhee different from Mem0, Letta, MemPalace, and agentmemory?

Accepted Answer

Dhee is the only agent memory layer that leads LongMemEval at R@5 99.4% on the full 500-question set, self-evolves its retrieval policy per tool and per intent, and ships a context router that digests Read, Bash, and subagent output at source instead of dumping raw output into context.

Question 4

Does Dhee work with Cursor, Codex, Gemini CLI, Aider, Cline, or Goose?

Accepted Answer

Yes. Dhee exposes an MCP server that plugs into any MCP-compatible client — Claude Code, Cursor, Codex, Gemini CLI, Cline, Goose, Claude Desktop — plus native Claude Code hooks, a Python SDK, a CLI, and Docker. One install, every host.

Question 5

Is Dhee production-ready? What storage does it use?

Accepted Answer

Dhee runs on SQLite by default — no Postgres, no Qdrant, no pgvector, no extra infrastructure. It has 1000+ tests, reproducible benchmarks committed in-tree, an MIT license, and works offline with Ollama embeddings or online with OpenAI, NVIDIA NIM, or Google Gemini.

Question 6

How does Dhee's self-evolution actually work?

Accepted Answer

Every time the model calls dhee_expand_result(ptr) to see the raw output behind a digest, Dhee logs the tool, intent, and depth. The command dhee router tune reads that ledger and atomically rewrites ~/.dhee/router_policy.json — deeper digests for file types the model keeps expanding, shallower ones it never does. No config to hand-maintain.