What did you learn from your AI coding session?

You sign in on your phone or any browser and read everything. Your archive lives in our cloud, encrypted at rest with AWS KMS, isolated to your account. There's no recovery code to remember and no device-pairing dance. Lost laptop, new laptop, new phone — you sign in, the archive is there. Two things require an extra step (re-confirming your sign-in, or a code from your authenticator app if you've set one up): exporting all your sessions as JSON, and deleting your account. Those are the irreversible / take-everything-and-leave actions, and we'd rather make sure it's actually you doing them. Everything else — reading, searching, capturing lessons — is one tap, every time.

A normalized copy of your AI session content — messages, code blocks, model names, timestamps — synced to your private workspace in our AWS account. Everything is encrypted in transit (TLS 1.2+) and at rest (AWS KMS), and isolated to your account at the data layer: there is no cross-user query path in our backend, by construction. A best-effort regex pass strips obvious secrets on your machine before anything is sent.

Yes — the half that runs on your machine. The VS Code extension and every provider parser (adapter) are MIT-licensed on GitHub. You can clone the repo, read what touches your local files, build the extension yourself, and verify it does what we say. The cloud backend and the mobile app stay closed-source — that's our work to maintain.

Obsidian is a great note-taking tool, but it doesn't know what an AI coding session is. You'd be pasting transcripts into notes that you never re-open. AgentsOfMine reads the structured session data your agents already wrote — bubbles, tool calls, code blocks, file references, model names — and renders them as a navigable history. Lessons are attached to the exact chat bubble that produced them, so the context comes free.

The free tier shows your last seven days. Older sessions are still on our servers — nothing is deleted unless you delete it — they're just outside the v1 UI window. Pro will unlock 14-day, 30-day, and all-time views when it ships. Until then, the seven-day default is everyone's window. JSON export covers everything regardless of the UI window.

OpenCode, Claude Code, and Codex are the providers we ship with. Each one has its own adapter — a small parser that knows that provider's local session format. The adapter contract is open and documented, and we run an explicit community-contributor program for adapters covering Cursor, GitHub Copilot Chat, Google Gemini CLI, Cline, Continue.dev, IBM Bob and others. If your tool isn't on the list, open an issue.

It's in active development. We aren't promising a date — we'd rather ship it right than ship it now. The waitlist is the only signal we'll send: one email, the week the alpha goes live, and nothing else.

Yes — eventually. The free tier ships with everything you need to read, search and capture lessons from your sessions — that's the product, not a teaser. A paid tier later will add things that genuinely cost money to run, like AI-generated session summaries, longer retention (sessions older than a few weeks), multi-device sync, deeper analytics across projects, and team workspaces. We won't lock features you already have, and we won't dangle locked features in the UI before they exist.

Yes — the highest-value path is writing an adapter for an AI agent we don't ship with yet (Cursor, GitHub Copilot Chat, Google Gemini CLI, Cline, Continue.dev, IBM Bob, etc.). The adapter contract is documented; the bar is a weekend's work for someone who knows the target tool. If your adapter ships in a release, you become a founding contributor: your name on this page and in the adapter's README, co-maintainer on the adapter's repo, and free Pro tier on us as long as you're maintaining the adapter (when Pro ships). No quotas, no first-N race, no leaderboard — the bar is shipping working code. The contributor guide is on GitHub.

Beyond lessons, two pillars are next: “what's still open” (auto-detected open/decided/blocked status on every session) and “what did this cost” (tokens per task, tool-call success rates, real model comparison on the work you actually do). No dates, no fake locked features. The full roadmap principles live in the project repo at docs/product/roadmap.md. Beyond those: making lessons actionable from the VS Code extension, making them collaborative, making them cross-project. None committed; same pricing principle as everything else — anything that genuinely costs us per-user money to run will be Pro, anything we can run cheaply stays free.