The last project
you'll ever re‑explain.
Agents start briefed.Decisions stay yours.Handoffs write themselves.
Brief and handoff in every tool. The mid-edit guard runs in Claude Code today.
claude plugin marketplace add Hark-Studio/hark-mod && claude plugin install hark-memory@harkhttps://harkstudio.io/mcpNothing is saved until you sign in. Every item links to the passage it came from.
Every session starts from zero. Not anymore.
Your teammate works in Claude Code, you're in Cursor, the client asks where it stands. Every person and every agent reads one short brief — what's decided, what's in progress, what not to retry — instead of re-reading the repo and guessing.
The code tells you what. Hark keeps the why.
A week away and the commits still show what changed. Hark holds the decisions behind them, the approaches that didn't work, and the next action — so you pick up where you stopped, not where you think you stopped.
Hand it to a person or an agent. They continue.
One link with the Current state, the decisions that still apply, and the open questions. When a session stops, Hark drafts the handoff for you — nothing ends without a note.
An agent working on V-004 can't read or write V-009 without an explicit switch — and blocked attempts are logged.
If a session goes quiet, Hark drafts the handoff from what actually happened; the next agent sees it first.
Decisions and state changes proposed by agents wait for your accept or reject; they never enter the brief or your public project page unreviewed.
Phase, next action, last decisions, blockers, non-goals — with the full project page on request.
With export on, Hark commits .hark/HANDOFF.md, .hark/DECISIONS.md, and an AGENTS.md block, so the brief survives even if Hark is unreachable.
Every entry carries the client that wrote it, so you can always see which agent said what.
From spark to shipped, in four moves.
Drop any idea in seconds. It lands in the Vault.
Business, build docs, branding — with a project page score that shows what's missing.
Move it through the pipeline: Vault → Recon → Blueprint → Fabrication → Live.
Publish a project page, open a client or investor room, or connect any agent through MCP.
Works with Claude Code, Codex, Cursor, ChatGPT, Grok — or no AI at all.
See the full walkthrough →Every idea gets a stage.
Move an idea from a raw spark to something live — an app, a newsletter, a service, a client build. Nothing gets forgotten in the middle.
The full stack of every idea, in one place.
Every project holds its business thinking, build docs, and brand assets — with a project page score that tells you what's still missing.
Thesis, problem, audience, model, competition, risks, next actions. The thinking that usually lives in a notes app finally has a home.
Approach, stack, scope, notes. Optional AI context doc for anyone building with Claude Code or ChatGPT.
Name, domain, palette, typography, logo drafts, and the files behind them — all attached to the idea, not floating in a folder.
Your ideas. Your team's. Your clients'.
Workspaces are separate drawers on one dashboard. Keep your personal side projects distinct from your co-founder's, and from the builds you're doing for someone else — without juggling accounts.
A private drawer for your own ideas — the ones nobody's seen yet.
Invite collaborators, assign roles, keep the pipeline in sync.
One workshop per client. When it's done, hand over the ZIP.
Pick your workshop.
Every new workshop starts with a free 14-day trial — no card. Solo starts on Builder, teams on Studio. Annual plans get two months free.
For the solo builder with too many ideas.
- Workshops
- 1 workshop
- Seats
- 1 seat
- Guests
- 1 client guest (comments, one project)
- Every project's decisions and next step in one place
- Any AI tool you use reads the same brief
- Hand off a project with one link
For small teams and founders working with others.
- Workshops
- 5 workshops
- Seats
- 3 seats · extra seats $12/mo
- Guests
- 10 client guests
- Teammates work on the same projects
- Comments and a history of who changed what
- Private rooms to show investors
For studios shipping for clients.
- Workshops
- Unlimited workshops — one per client
- Seats
- 8 seats · extra seats $10/mo
- Guests
- Unlimited client guests
- Each client kept fully separate
- Client pages with your branding, not ours
- Hand the finished record to the client's repo
For organisations that need it on their terms.
- Workshops
- Unlimited workshops
- Seats
- Unlimited seats
- Guests
- Unlimited client guests
- Sign in with your company login (SSO)
- Your contract terms and invoicing
- Priority support
Decision ledger: Which agent decided what, under which trust level, approved by whom — a record no single vendor can give you.
Limits — agent calls, storage, and what counts (reads are unmetered)
| Builder | Studio | Agency | Enterprise | |
|---|---|---|---|---|
| Agent calls that count / month | 10,000 | 50,000 | 250,000 | Unlimited |
| Reads (brief, lists, search) | Unmetered | Unmetered | Unmetered | Unmetered |
| Storage | 10 GB | 50 GB | 250 GB | 1 TB |
| GitHub connections | 1 | 3 | Unlimited | Unlimited |
A counted call is a write or AI-processing call an agent makes — logging a decision, moving a stage, ending a session, running an import. Reads are unmetered: the brief, start_session, lists, search, project context, repo activity, journal and checklist. A per-person burst limit still applies to every call. Anything you do by hand in the app never counts.
Building with AI? Plug Hark in.
Connect Hark to Claude Code or ChatGPT via MCP — capture ideas mid-session and keep specs and stages in sync as you ship. Optional; everything works without it.
Claude Code: install the Hark mod — starts briefed, ends with a handoff.
Your ideas are yours. We never train on your data, we never sell it, and we never hand it to third-party model providers for training. Read the Privacy Policy.
Shipping every week.
Reading the brief is free; briefs show their cost
Reads no longer count toward your agent-call cap — only writes and AI-processing calls do, and the usage graph shows the two apart. Every brief now ends with its approximate token count, what was left out, and how many items your repo's CLAUDE.md/AGENTS.md already covers.
Don't retry, rejected alternatives, and Resume in…
Record what was tried and didn't work; failed attempts show as a "Don't retry" list in the brief and handoff, and decisions carry the alternatives you rejected. Answers to recipient questions can become decisions, and "Resume in…" gives one copyable block for Cursor, Claude Code, Codex or VS Code. Hooks now fail open within 2 seconds, with a new hark doctor check.
Trust levels, Autopilot and one-click session approval
Each project picks how much agents approve on their own (Standard, Relaxed, or a 24-hour Autopilot per agent tool), marked Auto-approved everywhere. Decisions are never auto-approved, and anything that clashes with one waits for you. "Accept all from this session" clears the rest in one click.