v0.1.8

New

• fixed the single query refinement issue.
• Aligned interaction memory with LLM-Wiki (graphify). For example, as typed /deeprefine, it will automatically read the memory file of graphify and fetch the queries that are not marked with "true" in the deeprefine's own history file. Then deeprefine will refine the KB with them iteratively.
• Added new command: deeprefine history sync-memory -- Import graphify memory queries into history.

Next

• Do coverage-based greedy query selection.
• Adapt to more agent framwork.

v0.1.7

Agent-native DeepRefine loop for Cursor — same control flow as Reafiner.refine(), without FAISS or extra API keys.

New

• /deeprefine skill: multi-hop retrieve → → →
• Retrieval via graphify query + k-hop from graph.json
• deeprefine loop (init / validate / finish) + apply --trace-file
• agent_graph.py applies to graphify JSON
• Fix: CLI refine respects DEEPREFINE_EMBED_MODEL

Next

Support more agent platforms (Claude Code, Codex, …)

v0.1.3

The first deeprefine-cli skill release.

Highlights

• /deeprefine Cursor skill: install with deeprefine cursor install (project or --user);
• deeprefine CLI: history, refine, index, cursor install|uninstall;
• graphify integration: reads/writes graphify-out/graph.json; history under graphify-out/.deeprefine/;
• PyPI: pip install deeprefine-cli.

Typical workflow

Enter into a agent (now support Cursor):

• Build a KB with graphify: /graphify .;
• Ask questions: /graphify query "...";
• Record queries: /deeprefine history add --query "...";
• Refine the graph: deeprefine refine in CLI or /deeprefine in Cursor.

ToDo in Next Version

Now only support locally deployed vllm service. Need to support API services.