Local Mac workspace · bring your own Claude/Codex · evidence before claims

Show the work your AI
actually did.

Talkak is the cockpit around your own Claude Code/Codex sessions: panes, logs, summaries, Usage Pulse, WorkItems, blocked queues, and local memory so daily AI work becomes reviewable evidence.

14-day local app trial · Apple Silicon · paid checkout, notarization, and external connector claims stay gated until verified
Several of your agents, working at once — and the work still waiting on you.
Built in the open
1,177 commits · 130 dev logs · 111 post-mortems · 4 weeks · 1 founder
Every dead end is public — see the real build, not a highlight reel.
See the build journey →
See it work

Actual product screens, not mockups.

The cockpit, the seven-area map, your own agents in panes, the connectors, the assistant, and the approval gate — every screen here is the real running app.

Demo

Parallel agent work, recorded.

Real footage of multiple Codex agents working across projects at once. The value is not the spectacle; it is keeping the work visible enough to review.

What Talkak does

It turns scattered AI sessions into reviewable work.

Talkak wraps your own Claude Code/Codex with a local operating layer: multi-pane sessions, transcript logs, summaries, Usage Pulse, WorkItems, and guarded approval/receipt flows. It does not call AI or mutate external providers for you.

Run every business from one screen

Every company, agent, and waiting decision — in one cockpit.

See which agents are online, which businesses are live, and the exact count of things waiting on you. Then click straight into the one that needs a decision.

  • Each service surfaces its “needs you” first — no digging through terminals to find what's blocked.
  • Built for a solo founder running many things at once, not one app at a time.
One shape for any business

Every service splits into the seven areas that run it.

Build, Lab, Run, Reach, Sell, Earn, Guard, Steer. Company or personal project, the same seven areas hold the work — with Connect, Weekly report, Billing, and Backup one click away.

  • Entity → Service → Area → widget: the same hierarchy, top to bottom.
  • Areas stay industry-agnostic, so a café and a SaaS use the same map.
Bring your own agent

Run your own Claude Code / Codex — many sessions at once.

Keep a terminal pane per project and run several Claude/Codex sessions in parallel, each where its work lives. A code drawer shows the diff, and every session keeps a re-openable conversation log and summary. We wrap your agent — we never call AI for you or mark up a token.

  • Multiple agents side by side — one pane per startup or repo, switch in a keystroke.
  • Re-open any session's full conversation log and summary; recover context after a restart.
  • Native Rust + Tauri core — lightweight and keyboard-first.
Operating memory

Connect your real tools — and remember everything.

Pull in Gmail, GitHub, Stripe, Calendar and more. Talkak builds a local memory of what happened across every project, and shows it as a connective graph of changes you can actually search.

  • You connect what you want — nothing is required, and sources stay local.
  • Ask later and it answers from your real history, not a guess.
Ask the assistant

Ask anything about the business — answered from what actually happened.

“What did we decide?” “What shipped this week?” The assistant answers from your sessions, commits, and connected sources — and points back to the receipt, not a hallucination.

  • Every answer traces to a real session, commit, or message.
You stay in control

AI drafts the reply. You press approve — or it never sends.

Email replies, calendar events, and doc edits are staged as drafts. Nothing leaves your machine until you approve it, and every approved action keeps a receipt.

  • External, hard-to-undo actions sit behind an approval gate by default.
  • No silent sends, no surprise edits — proof before claims.
More than a wrapper

Your AI, with a control layer on top.

Talkak wraps the Claude Code running in each pane — same model, but it lies less and remembers more. Every turn passes through verification, memory, and an approval gate.

Re-run, not trusted

“Tests pass.” Talkak checks.

When the agent claims a test passed or the build is green, Talkak re-runs that exact command itself, records the real result, and feeds the correction back to the agent if it was wrong — a false “done” gets caught, not believed.

Every turn, a recap

⌘I — a plain-language card.

Hit ⌘I on any pane and see what the agent just finished, what changed, and the next step — read from the session transcript, not the scrolling terminal.

Work becomes memory

Decisions remember themselves.

The durable facts and follow-ups your agent flags are captured into your private project memory automatically — no setup, and it works on a clean install.

Nothing fires silently

Risky actions wait for you.

Email, calendar, and doc actions stage behind an approval gate, and an honesty directive rides in every pane's Claude — proof before claims, by default.

Verification, summaries, and memory apply to the Claude Code in each pane. Codex panes get project memory; the rest of the layer is Claude-first.

Why "Talkak"

The name is the sound of it working.

“Talkak” comes from the Korean sound 딸깍 — the little click of switching something on, or of a thing getting done. That's the whole product: the work happens, and you hear the 딸깍.

Built in the open

The real build — not a highlight reel.

Four moments from 1,177 commits in four weeks. The honest failures, the hard systems, the product actually working — pulled straight from the dev logs.

See all 130 dev logs →
Pricing

BYO workspace now. Paid checkout after owner verification.

Talkak is the workspace — plug in your own Claude Code / Codex and we never mark up a single token. Paid launch terms stay gated until price, Merchant of Record, legal copy, and license flow are owner-confirmed.

Paid checkout pending
Local trial is available now; paid subscription terms are not launch-ready.
  • 14-day local app trial in the desktop build
  • Usage Pulse, ⌘I summaries, the connective graph
  • 0% token markup — bring your own Claude / Codex
  • Checkout, provider, legal, and license facts require owner evidence before paid launch
Download the current Mac build
Paid checkout is disabled in public copy until the owner checklist is complete.
Download

Get Talkak

Talkak app icon — the sloth-robot mascot in a rounded-square macOS icon frame.
⬇ Download for macOS (.dmg) macOS · Apple Silicon · release metadata still gated
🍎 macOS 13+ 💻 Apple Silicon (M1+) 🤖 Your own Claude Code
Release gate pending. Signing, notarization, updater, and first-open behavior must be verified against the exact distributed artifact before public launch claims.