Hark gives every idea one home base: the plan, the live state, the agent brief, the team record, and the proof you choose to share.
A private space for a set of projects — yours, your team's, or one per client.
Keeps unrelated work from mixing; the boundary agents and permissions respect.
Six stages every idea moves through: Vault → Recon → Blueprint → Fabrication → Live → Cold Storage. Shown when you turn on Advanced structure; the simple view keeps to Project, Current state, Decisions and Next action.
You can see at a glance where every idea stands, and moving one is a real decision.
Each idea is a structured file: Business, Build Docs, Branding.
The full stack of an idea lives in one place instead of chats, repos, and folders.
A completeness score from a checklist you control.
Shows what's still missing before an idea is ready to build or share.
A one-line inbox for raw ideas, promotable to a full project later.
Capturing takes seconds, so ideas stop dying in group chats.
Paste a ChatGPT or Claude thread, connect a repo, or point at an existing project — Hark proposes a Current state with quotes as evidence, and says "Not found in the source" instead of guessing.
Rescues the ideas already buried in dead chats — you start with work, not a blank page.
Paste a conversation or tap the sample and see a proposed Current state before signing up; it carries into your account.
You see the result in seconds before committing to anything.
A drag-and-drop view of the pipeline.
Moving ideas between stages takes one motion.
Find any idea across workshops.
Works at 4 ideas or 400.
Leaving the Vault requires a next action, a kill criterion, and a review date.
Forces honesty about each idea before it consumes attention.
Parked ideas live in Cold Storage; stale ones get flagged during review.
Your active list stays real; nothing quietly rots.
A short, always-current card: phase, what shipped, in progress, next up, blockers, open questions, non-goals.
The one screen that answers "where exactly are we?" — for you and for any agent.
A dated log of ships, decisions (with rationale), milestones, and notes.
Your history survives; the why behind every choice is written down.
Each work session ends with a note: what changed, what's next, what to watch.
The next person or agent picks up exactly where the last one left off.
Every record shows its source (person, agent, repo, imported chat or doc), its evidence, whether it's accepted, and whether it's still current.
You can tell a confirmed fact from an agent's claim at a glance.
Commits, merged PRs and releases become an "Inferred from repo activity" draft with each commit linked as evidence.
Progress gets recorded even when nobody wrote a note.
Define metrics (MRR, users, signups), log values, see sparklines.
Numbers tell the story — and become the investor update.
Repo, live site, staging, app stores, socials, docs on every venture.
The project connects to the real thing.
After launch: a feature pipeline (idea → shipped) and a feedback inbox.
The roadmap keeps living after the product does.
Branch an idea into a pivot, spinoff, or v2 with recorded parentage.
Ideas become a family tree; pivots keep their history instead of losing it.
Connect private repos; releases and merged PRs land in the ship log automatically.
Proof of progress comes from the code, not from memory.
A live-site preview card and the repo README on the venture.
Every idea gets a face without extra work.
Claude Code, Codex, Cursor, ChatGPT, Grok, VS Code, Windsurf, Gemini, or your own agents connect to Hark.
One shared memory across every tool — models change, Hark persists.
One call gives an agent everything it needs: state, decisions, blockers, non-goals. Short by default, full on request.
Any agent is fully onboarded in one read, without skimming a bloated file.
Start a session, work with tools, end with a handoff.
Turns agent work into something the next agent can continue.
If a session goes quiet, Hark drafts the handoff from what actually happened.
Nothing ends without a note, even when nobody wrote one.
A session is bound to one project; touching another requires an explicit switch, and attempts are logged.
Context never leaks between your projects.
Decisions and state changes proposed by agents wait for your accept or reject, reviewed as a diff with Supersede or Keep both.
Agents can propose; only you make something true.
Per project: Standard reviews everything; Relaxed also auto-accepts list items like Next up and repo drafts; Autopilot (one agent tool, 24 hours, with a banner) accepts everything except decisions, non-goals and phase. Auto-accepted items carry an Auto-approved badge everywhere, including handoffs.
Less clicking on routine updates, while decisions — and anything that conflicts with one — always wait for you.
One click accepts every quoted or observed item from a session; decisions, non-goals and phase changes stay as a short list to decide individually. Also linked from the daily digest.
Reviewing a whole session takes seconds, not a scroll.
Accepting a new decision can retire a contradicting old one, with history kept.
The brief reflects what you believe now, not everything you ever said.
Every agent write shows which agent did it: "via Claude Code", "via Codex".
You always know who changed what.
Long-lived tokens for scripts, workflows, and custom agents. Read-only tokens can read the brief and project context but can never write.
Headless automation gets in without a browser.
Step-by-step setup for every supported client, including Grok, copyable as Markdown.
Any builder — or their agent — connects in two minutes.
hark init writes hooks and rules into the repo so agents read Hark automatically; in Claude Code and Cursor, stopping a session drafts the handoff.
Consulting Hark becomes automatic, not remembered.
Install buttons for Cursor and VS Code, with a web-link fallback.
Connecting an editor takes one click instead of a config file.
Optionally use the repo's CLAUDE.md as the source. Hark compares both copies before syncing, warns when the repo copy is empty or much shorter, and keeps a version history you can restore.
Turning sync on can never silently wipe the context you wrote.
Hark commits HANDOFF.md, DECISIONS.md, and an AGENTS.md block into your repo.
The brief lives in git too — it survives even if Hark is unreachable.
A graph of agent calls by client, with your plan's cap and warnings at 80% and 100%.
No surprise limits; you see what counts and when.
Hand off to a developer, client or agent: a gap check flags what's missing, the recipient can ask questions, and each handoff is versioned. Recipients can add proposals without a seat or workshop access.
The next person continues the work without rebuilding context.
Owner, admin, member, and venture-scoped guests.
The right people see the right things.
Read-and-comment access to specific ventures for clients, advisors, reviewers.
Feedback without handing out edit rights or seats.
Discussion on any project, live.
Decisions get made where the work is.
Everything that happened, by whom or which agent.
Full visibility for the team.
A two-minute screen: confirm next actions, resolve questions, accept agent proposals, park stale ideas.
The habit that keeps every idea honest.
Slack, Discord, Telegram, and email on ships, stage moves, decisions.
Hark shows up where your team already lives.
Link a group chat to a workshop with /link; @mentions and /spark become sparks, /v adds a journal note, and chosen updates post back to the group.
Team chat feeds the project without anyone copying things over.
Message the Telegram bot or email your workshop address — it becomes a spark.
Ideas get captured the moment they happen.
A checklist that completes itself as you use the product.
New users learn by doing, not by reading.
Every workshop is isolated at the database level.
Your ideas are never visible to other customers.
Hark never trains on your content; export or delete anytime.
You can trust it with unlaunched ideas.
Standard, secure sign-in; email accounts confirm their address first.
No new passwords to manage.