Changelog

What changed, in plain words. This file ships with the engine — the dashboard's Changelog tab and whatever-recall.com/changelog render exactly this text, so the tool and the site can never tell different stories.

Versioning follows semver: MAJOR.MINOR.PATCH. Anything that could break your existing .mind/ index or your workflow lands only in a MAJOR release and is called out explicitly.

Coming next — the roadmap

We alternate build releases with harden releases — ship a layer, then stabilise it on real feedback before the next. Nothing below is shipped yet, and we'll always say so plainly.

• v1.1 — hardening. Bug-fixing and polish from real-world use. No workflow changes; your .mind/ index stays exactly as it is.
• v1.2 — the app & deeper features. A real native app (not just the local dashboard), with more: write-side features, real project planning, and a proper team wiki.
• v1.3 — hardening. Stabilise the app on feedback from real teams.
• v1.4 — structure as a render (the horizon). Today your code lives in folders and files — a structure built for human reading. An AI navigates by the graph, not the tree. recall already holds that graph (every dependency, every decision, every link). This step turns it around: the graph becomes the truth, and the folder/file layout becomes a view rendered from it — so the AI can organize and design code the way it actually thinks, while recall renders it back to ordinary files for your compiler, your team and git. The same move as search-inversion, one layer deeper. The deepest change, taken only when the ground under it is solid.

v1.0.3 — your notes keep their own identity

A fix and a small feature, both about one thing: a note is its own note.

• Fixed — two different notes on the same file no longer merge into one. When you stamped a second decision onto a file that already had one, recall could fold it into the first and lose what you just wrote. Now a merge only happens for a real re-statement (the same note, said again) — a new title is a new note, even on the same file. Nothing you write gets quietly swallowed.
• New — recall stamp --id <n> edits exactly one note. Like editing a task by its id instead of its name: same name allowed across notes, the id is the identity. Fast, unambiguous, no guessing which note you meant.

v1.0.0 — whatever-recall is live 🎉

This is the launch. whatever-recall is now publicly available — give your code a memory, and let any AI read it back in milliseconds at zero tokens. Everything below is in this first release; nothing here is "coming soon".

Start here: Install · Docs — the full guide · How it works · Pricing · Source on GitHub · Report an issue

Here's everything you get on day one.

The engine

• recall init . builds your project's memory locally: code map, decisions, lessons, commits and tasks in one SQLite file inside your repo (.mind/). No cloud, no telemetry — delete the folder and it's gone.
• Reading costs zero tokens. Recall answers from the index in milliseconds, with three parallel tracks: the relevant code (ranked by importance), the knowledge behind it, and the blast radius — what breaks if you change it.
• The memory keeps itself fresh. Git hooks stamp every commit; the watcher re-indexes new commits on its own; drifted knowledge is flagged, never silently trusted.
• Power Mode (optional): connect your AI (Claude Code CLI, Ollama, an Anthropic key, or any OpenAI-compatible endpoint) to enrich the index with semantic summaries. With a subscription CLI it costs nothing extra.
• Search inversion (recall resolve): an AI often guesses the wrong search term from its training. recall corrects the term into your repo's real vocabulary before anything is grepped — the moment everyone else ignores.
• Static code intelligence — token-free, offline. recall callers / callees (who depends on a file, and what it depends on), impact (what a change touches: empirical co-change + structural dependents — a 0-token stand-in for call-hierarchy), precedent (the most analogous past decisions, with how each turned out), plus dead-code, untested and cycles candidates — all read straight off the index, no model, no tokens.
• Private knowledge stays local. The why behind your code lives in the local brain (.mind/), never in your source — so your code ships clean by construction, with nothing to scrub before you publish. Every note carries a visibility: team (shareable) or private (recall stamp "…" --private, yours alone). recall export writes a shareable brain with all private notes stripped, and two fail-closed guards make it physical — the export aborts rather than leak, and recall check-leak blocks a commit that would stage a private brain. Your decisions never leave your machine unless you choose to share them.

The dashboard

• recall dashboard — the browsable wiki: Overview, Wiki (causal chains down to the diff), Tasks, Git, Drift, Code (with the pre-edit briefing), Product tree, Search, and this Changelog. Live pulse, one design system.
• Tasks police their own status. A task with every step ticked but still marked open gets flagged — in the dashboard and as a one-line nudge on every commit — until you flip it to done. Your backlog can't quietly rot.
• recall shortcut — a desktop launcher (with the recall icon) that starts the dashboard with a double-click. No terminal, no AI needed. It runs the dashboard in the background (recall tray — a system-tray icon with the [tray] extra, otherwise a console that loudly warns don't close me), so a closed window can't silently take the server down. recall stop ends it cleanly. And if the server ever blinks (a restart), the dashboard's live pill reconnects in about a second instead of hanging.
• GitHub and Discord are one click away — icons in the dashboard header link straight to the repo and the community.
• Built with — the About tab credits the stack recall stands on (Vercel, Supabase, Claude Code, Next.js, React, TypeScript, Python, Stripe, Resend), with offline inline logos.

The doors for your AI

• MCP server (recall mcp) — 14 native tools in Claude Code, Cursor and any MCP client: recall, brief, explain, resolve, stamp, contested, freshen, dashboard, impact, precedent, callers, dead_code, untested, cycles.
• CLI — recall "<question>", brief <file> (read before you edit), explain (onboarding), review <commit>, contested, precommit-check, plus the code-intelligence commands (callers, callees, impact, precedent, resolve, dead-code, untested, cycles). recall --version reports the installed release.
• Built for agents and hookless harnesses. recall push gives a subagent the same task-scoped brief + landmines + live 🔴 BROKEN warnings the pre-edit gate would show, over the shell; recall ack <file> confirms a briefing so the gate lets the edit through. recall sync-context writes recall's live state into your AI instruction file (CLAUDE.md / AGENTS.md / Copilot) so every client loads it with no tool call. recall receipt reports what recall actually did over a window — in measured counts, never an invented token or dollar figure.
• rules.md — the transparent, downloadable contract that tells an AI agent exactly how to use the memory. No hidden prompts.

Accounts & licensing

• 14-day full trial, no card. Sign up with email or Google / GitHub / Discord. One plan — Developer, every feature included; you only buy seats: $29/seat·month or $290/seat·year (2 months free yearly), minimum 1 seat.
• Offline license tokens (Ed25519): the tool verifies your license without ever seeing your code — when a trial ends, reading your memory keeps working; stamping new knowledge needs a plan.
• The full source stays public on GitHub (Business Source License 1.1) — read every line that touches your repo before you run it. Noncommercial use is free; commercial production use needs a plan; and every version becomes Apache 2.0 open source three years after its release.