v0.3.1

• README: LongMemEval_S retrieval results for the local install, 91.6% Recall@5, 94.4% Recall@10 and 0.915 MRR over 470 questions, with each question type and the 30 abstention questions scored on their own.
• README: real logos for the 22 agents that showed letter placeholders in the "Import directly from" grid, and an "Any MCP agent" tile.

v0.3.0

• Step-by-step terminal UI. make0-brain typed in a terminal opens a menu, and commands run without their arguments ask for them one step at a time, with arrow keys, checkboxes and a summary before anything changes (import, import --sessions, init, brains create|edit|delete, agents grant|revoke, export, restore, purge, history). Built on @clack/prompts. Arguments, --yes, or no terminal skip the steps, so scripts and MCP clients behave as before.

• import --sessions saves one memory per past coding-agent session, in a category per project, read from the history 41 agents keep on disk: Claude Code, Codex, Cursor, VS Code Copilot Chat, Copilot CLI, Gemini CLI, opencode, Cline, Roo Code, Kilo Code, Continue, aider, Zed, Goose, and more (import --list-agents lists them). Each one records what you asked, your corrections, the files that changed, the commands that failed, and how it ended. It needs no model or key. Credentials and long random tokens are redacted before anything is stored. It reads the last 30 days and at most 500 sessions by default (--since and --limit, each taking all), and never opens a session outside them. --agent (a comma-separated list) narrows it down, and with --extract a model keeps only the facts that stay true after the session.

• Local models with no key. A localhost MAKE0_BRAIN_LLM_BASE_URL (Ollama, LM Studio, llama.cpp) turns on duplicate merging, categories, and --extract, and nothing leaves the machine. Local calls get longer timeouts, search stays on the local reranker unless MAKE0_BRAIN_LLM_SEARCH=1, and servers that reject JSON mode are asked again without it.

• MAKE0_BRAIN_BASE_URL is now MAKE0_BRAIN_LLM_BASE_URL, so the name says it is the language model's endpoint. The old name still works.

• README: an "Import directly from" section with the logos of the 41 agents, terminal screenshots of the menu and of a session import, and a coherence pass over the rest.

v0.2.4

• Requires Node.js 22 or newer. better-sqlite3 13 needs Node 22, so Make0 Brain crashed on Node 20 even though engines allowed it. engines, the README, and CI now say Node 22. Node 20 reached end of life in April 2026.
• README: npm, license, Node, CI, mcprush, and Glama badges; one-click install buttons for Cursor and VS Code.
• CI runs the typecheck, tests, and build on Linux and macOS with Node 22 and 24.
• Releases publish from GitHub Actions with npm provenance, and update the MCP Registry and GitHub Releases from a tag push.
• SECURITY.md, CONTRIBUTING.md, this changelog, and issue templates.

v0.2.3

• init detects Hermes Agent (~/.hermes/config.yaml, mcp_servers) and OpenClaw (openclaw mcp set, or mcp.servers in openclaw.json).
• The viewer shows Hermes Agent and OpenClaw logos.
• README: mcprush badge.

v0.2.2

• Security: the viewer answers only its own Host (127.0.0.1:<port> or localhost:<port>) and refuses writes from other origins or without a JSON content type. Before this, any page open in the browser could send simple cross-origin writes, and DNS rebinding could read memories.
• README: a "Verify it works" walkthrough, a Privacy and security section, and Backups, upgrades, and support.

v0.2.1

• The viewer's search box runs the same hybrid retrieval as the MCP server (semantic + BM25, rerank, graph expansion) instead of filtering the list in the browser.
• BM25 matches word prefixes, so a partly typed word still finds the memory.
• README: local search takes about 80ms per query on an Apple M2 Pro.

v0.2.0

• Brains: create, edit, and delete custom brains alongside personal. brains and agents control which app sees which brain, and import, export, view, and stats take --brain.
• Usage: per-app, per-brain daily reads, writes, and token estimates, on the viewer home page and in a Brain activity chart.
• The viewer is rebuilt in React, with a board and a 3D memory graph.
• Listed in the official MCP Registry and on Glama.

v0.1.0

• First release: one versioned memory for Claude Code, Claude Desktop, Cursor, Codex, Windsurf, VS Code, and Gemini CLI, served over MCP from a local SQLite file.
• Hybrid search (MiniLM + BM25 + cross-encoder, two-hop graph expansion), per-write versions with writer attribution, archive and restore.
• init for MCP clients; import from Markdown, Obsidian, and agent rule files; Markdown and JSON export; a local viewer; optional bring-your-own-key LLM features.