v0.2.0 — one shared brain for every AI

0.2.0 (2026-10-09)

hamyad is now one shared brain for every AI tool, not just Claude.

Connect any AI

• hamyad init --all (or --tools …) wires MCP + hooks + instructions for Claude Code, Codex, Gemini CLI, Cursor, Copilot (VS Code, CLI, cloud agent), Windsurf/Devin Desktop, Zed, Roo Code, JetBrains Junie and aider; --global adds Claude Desktop, Codex user config and Windsurf's global MCP.
• hamyad connect <tool>: exact setup for ChatGPT (MCP connector and Custom GPT Action), claude.ai, Claude Desktop, Grok, Perplexity, Gemini app, Cline, JetBrains AI Assistant and "anything else".
• REST API + OpenAPI 3.1 (/api/*, /openapi.json) on hamyad serve and the Worker, for GPT Actions and tools without MCP; GET /api/memory.md for tools that read a URL.
• ChatGPT gets OpenAI's read-only search/fetch tools (deep research compatible). Writes are attributed per client (?source=, MCP clientInfo, User-Agent).

Capture across tools

• One hook entry point, hamyad hook <tool> <event>, for Claude Code, Codex, Gemini CLI, Cursor, Copilot and Windsurf (payloads verified against the real CLIs). Fail-open; errors go to a local log.
• Every agent session becomes a sessions/ entry and its git diff a changes/ entry (change journal; working-tree snapshots, generated files excluded, dibs attribution when present).
• hamyad import: transcript importers for Claude Code, Codex, Gemini CLI, Copilot CLI, Cursor agent transcripts, Windsurf and aider.
• hamyad timeline: chats, change sets, decisions, supersessions, tasks, agent-attributed git commits and dibs edits.

Decisions that change

• supersedes on brain_remember / --supersedes / hamyad supersede; superseded entries are linked both ways and listed under "Superseded — do NOT follow".
• Conflict hints when a new decision overlaps one still in force.
• Per-tool "last seen": every agent gets "⚠ DECISIONS CHANGED since your last session" at session start, and a mid-session notice on the next prompt.

No-GitHub mode

• hamyad init --store PATH keeps the brain in any synced folder or a single MEMORY.md file (Dropbox, Google Drive, iCloud, a cowork folder); .hamyad.json points the repo at it.
• .brain/MEMORY.md export regenerated on every write (CLI, MCP, hooks, Worker), re-importable; hamyad export.

Triggers and privacy

• Documented triggers: session start, prompt, turn end, session end, MCP/REST write, git post-commit, hamyad watch timer, manual sync.
• Secret redaction on everything written; capture.sessions off | summary (default) | full (opt-in redacted transcripts); size limits; per-tool switches; excludes.

Tests: 54 unit/integration tests plus scripts/e2e (46 checks) running the real Claude Code, Codex, Gemini CLI, Copilot CLI and aider binaries against a mock model server.

Breaking: hook commands are now hamyad hook claude SessionStart etc. (hamyad hook session-start|session-end still work; re-run hamyad init to update). Session records moved from .brain/.state.json to .git/hamyad/.