Hark keeps one clear record of each project: where it stands, what's been decided, and what happens next. Bring in what you already have, confirm what's true, then hand it off — to a developer, a client, or an AI agent — so they can pick it up without you re-explaining everything.
Bring in your existing work in one of three ways: paste a conversation you had with an AI tool, connect a GitHub repo, or start blank. Hark reads it and suggests what it found — decisions, the next action, open blockers — each linked to the exact passage or commit it came from.
Nothing is filed until you say so. Where the source doesn't say something, Hark shows "Not found in the source" instead of guessing.
Every project has one short answer to "where does this stand?": what's done, what's in progress, the next action, and anything blocking it. Anything a person hasn't confirmed yet is marked "Proposed" with a Confirm button, so you always know what's agreed and what's a guess.
With a repo connected, merged pull requests and releases show up as things that actually happened — useful when no one has written an update in a while.
The Decisions list shows only what's been accepted. Suggestions — from an import, an agent, or a teammate — wait in a review list until someone accepts or rejects them, and they never look like decisions in the meantime.
When a decision changes, the new one replaces the old, and the old one stays visible as a one-line history — so nobody works from an outdated call. You can also keep both if they genuinely both apply.
When someone else needs to pick up the work, choose who it's for — a developer, a client, or an agent. Before you send, Hark checks for gaps (a decision with no reason, no next action) and lets you fix each one in a click.
The recipient gets a link. They can read it for free, ask questions you answer in place, and add their own notes as suggestions for you to review — without taking a seat on your plan. Client handoffs only ever show confirmed, client-facing items. You can revoke a link any time.
Connect Hark to the AI tools you already use — Claude Code, Codex, Cursor, ChatGPT, Grok, VS Code, Windsurf and more. Most take a button or one line. Once connected, every tool reads the same current state before it starts and leaves a note when it's done, so the next one picks up where the last left off. It's optional — Hark works fine with no AI at all.
Set up your tools → · See agent features → · MCP server for agent memory →
A handoff only helps if it actually gets written. Five things make sure it does:
If you run many projects at once, turn on Advanced structure from your user menu. It adds a six-stage board for moving ideas from rough notes to shipped:
It also adds a fuller project page (business, build and branding notes), links, numbers, and an activity view. See the features page for details. You can turn it off again any time; nothing is lost.
Every new workshop starts with a free 14-day trial and no card — solo workshops on Builder, teams on Studio. Hark is named for the Harkness table, where every seat is equal and every idea gets heard.
Every record shows Agreed / Observed / Proposed and a source you can open: where did this come from? Agreed means accepted, not independently verified. Observed is reserved for system evidence such as a merged pull request. An agent's claim is never shown as verified simply because it was accepted.
Learn more about Evidence on every lineA resumed handoff means a different tool picks up the same project within seven days. Hark currently measures a different-tool session after a clean or confirmed handoff, not every brief read. Clean-handoff rate tracks sessions ended with a handoff; time-to-stale tracks the age of dated project state. Recipients answer ‘Was this handoff complete?’ and mark missing blocks. The owner adds gaps as open questions, keeping decisions Proposed. Two handoffs missing the same block trigger a warning before sending. These are how Hark judges itself, not a promise that every handoff succeeds.
Learn more about Measured continuityNew ventures, internal work, content programs and agent teams start Relaxed. Client builds and existing products start Standard, with every agent change reviewed. Existing projects keep their setting. Per-key trust overrides project trust. Decisions, non-goals, phase changes and decision conflicts interrupt you; ordinary notes and progress can auto-accept at Relaxed. ‘Accept all from this session’ groups eligible records. ‘Auto-accepted today’ has undo. After five decisions from one key on one project in a UTC day, extras become flagged notes.
Learn more about Quiet by defaultChoose a daily time for decisions waiting. Telegram has Accept / Reject buttons. Slack nudges link back to review, rather than offering Slack approval buttons. With ‘May relay my approvals’ enabled, your own agent can record your verbatim answer, channel and timestamps as evidence. It cannot turn its preference into your approval or re-accept something you rejected.
Learn more about Approvals where you already areOnce linked and selected as the source, the repo's CLAUDE.md or AGENTS.md is authoritative. Hark mirrors it on pushes and retains history rather than authoring competing edits. Optional .hark/ export carries accepted handoff state and decisions into the repo. Repo activity suggests drafts with commit evidence. Agent briefs include mirrored instruction documents and linked repo metadata. Arbitrary repo-document browsing is not exposed by the current tools: Hark never serves source code. Automatic session hooks and compaction protection in the Hark mod are Claude Code-only; other tools follow their connection guide and explicit session tools.
Learn more about Repo-native rulesHark reconciles planned work against repo facts on connection, incoming activity and nightly. ‘Looks done’ lists evidence-linked work for review, not a claim that code was tested. Tasks reversing decisions get conflict flags; tasks implementing them do not. Review stale items as done, still relevant or removed. Summaries show when they were written, and automatic summaries never silently overwrite human text.
Learn more about A plan that stays trueNew agent keys default Propose: suggest changes, never approve them. An eight-character pairing code expires after 15 minutes and works once; no key is pasted into a setup chat. One-link agent approval asks you to sign in, choose permissions, projects and expiry, then approve. The agent still completes pairing. Keys normally expire after 90 days and can be revoked or rotated. The separate show-once Copy key path is for private storage, not chat.
Learn more about Scoped, expiring agent accessInventory holds where work and access live, never credentials. Health holds evidence of what's working, blocked or uncertain. Work holds Tasks and next actions. Authority holds owner rules, goals and limits. Decisions hold choices with reasons and sources. Briefs, Memory and handoffs read the same record, answering: What is this? Where is the work? What condition is it in? What must I preserve? What am I responsible for next? Client views omit private access details.
Learn more about One five-block project recordAgent teams show labeled agents, who checked in, Tasks with exclusive claiming, a dated feed and a task briefing from the same project. Ready for a task does not mean actively working. Notes and completions follow trust settings; decisions always stay in Needs you for one-at-a-time review, with bundled asks when several are waiting.
Learn more about Labeled agents, shared Tasks