NewLaika Orbit recall for Claude Code: answers with zero model calls

Mission control
for your AI agents

Run Claude Code agents side by side, with a real browser, your inbox and calendar, and a live map of everything you know. One window, on your own Mac. Open source.

Get started for free Star on GitHub

Already in Claude Code? Add Laika Orbit recall with$ npx laikaorbit mcp

The Laika Orbit app window: a navigation rail, a calendar and an inbox of three agents waiting for you, a knowledge map of the indexed workspace grouped into repos, mail, chats, notes, documents and workspace files, and the Claude panel with two chats side by side, one working through a task list and one asking a question.

laika·orbit

⌕Ask anything…⌘K

recalllayoutARMS▾?

indexed327files275links8curated

work

5h42%wk61%

personal

5h11%wk23%

decisions.mdharbor-web/docs2.4 KB · 6 links

live6 groups · 4 rings

327 nodes · 275 linkstracking · 2 sessions

laika-orbit-site2atlas-apiharbor-web+

>_×

Passkey sign-inW×Seed data+

Add passkey sign-in to the admin console, behind the admin_passkeys flag.

Readsrc/auth/session.ts

Editauth/passkey.ts+48 −3

41export async function verify(req) {

+ if (!flags.passkeys) return null

+ const cred = await find(req.id)

− return legacyLogin(req)

BashRun the auth tests12 lines

INpnpm test auth --run

OUT✓ 14 passed (1.9s)

Registration and login pass. Wiring the email-link fallback next, then the sign-out on first passkey.

✦Computing…(1m 12s · ↓ 2,418 tokens)

▾Tasks 3 of 5 · Wire the email-link fallback

Passkey registration endpoint
Login with WebAuthn assertion
Tests for both flows
Wire the email-link fallback
Sign out old sessions once

Queue another message…

⚡ Auto ▾Opus ▾@/work ▾

Checkout auditP×+

Greparia-describedby9 files

ReadStepIndicator.tsx

EditCardFields.tsx+22 −9

Fixed 11 of 13 issues: focus order, labels on the card fields and contrast on the error text. The last one needs a call from you.

Question · waiting 26m

The step indicator has no text equivalent. How should it read?

1Visually hidden text“Step 2 of 4”, read out, no visual change

2aria-current on the stepLighter, leans on list semantics

3Leave it for designNote it in the PR

AnswerSkip1–3 · ↵

Suggested nextOpen the PR as a draft×

Reply to Claude…

⚡ Auto ▾Opus ▾@/personal ▾↑

Runs Claude CodeMCP nativemacOSMIT licensedLocal-firstNo telemetry

The problem

Your editor runs an agent.
Your day runs on twelve windows.

Editor · agent panel

zsh — claude

zsh — claude (2)

Chrome · Work

Chrome · Client

Mail

Calendar

notes.md

Usage · Settings

Today An editor, six terminals, three browser profiles, mail, calendar and notes, and you as the glue.

Laika Orbit

Needs you · 314:00 Design reviewWork 62%

With Orbit Your agents, browser, inbox, calendar and knowledge in one window, in orbit around the work.

Agents

Many agents. One view.

Run Claude Code in four chats at once, each in its own repo or across a whole project folder. Split them like editor groups and see at a glance which one is waiting on you.

Laika Orbit's Claude panel with four Claude Code chats in a two by two grid: one asks how failed webhooks should be retried, one is running a checkout test, one has finished an accessibility audit with all 13 checks passing, and one is writing retry docs against a three-item task list.

Add exponential backoff to webhook retriesatlas-apiRate-limit the export API

Add exponential backoff to webhook retries. Slow endpoints are piling up in the queue.

Readsrc/webhooks/deliver.ts

Retries run on a fixed 10s loop today. One choice before I change it:

RetriesHow should failed webhooks be retried?

1Exponential backoff1s, 2s, 4s… capped at 5 minutes, with jitter2Fixed delayEvery 30s, up to 10 attempts3No retriesLog the failure and move on

Reply to Claude…

Edit automaticallyOpusWork

Fix flaky checkout testpocket-app

Fix flaky checkout test

It fails when the payment mock resolves after the cart timer fires. Running the suite on its own first.

BashRun the checkout tests

npm test -- checkout

OUT

PASS  src/cart/cart.test.ts (2.1 s)
RUNS  src/checkout/checkout.test.ts
  ✓ applies a promo code (412 ms)
  ✓ keeps the basket after a failed payment

✻Testing…41s · ↓ 1.8k tokens

Queue another message…

AutoSonnetPersonal

Audit form accessibilityharbor-web

Audit form accessibility

Thought for 8s

Editsrc/forms/SignupForm.tsx+6 −2

BashRun the axe checks

✓13 of 13 checks pass

All three forms pass now. Every field has a label tied to its input, errors are announced through an aria-live region, and focus rings meet 3:1 contrast.

Reply to Claude…

Edit automaticallyOpusWork

Document the retry scheduleatlas-api

Document the retry schedule

Thought for 4s

Readsrc/webhooks/retry.ts

Failed deliveries retry after 1s, then double each attempt up to a 5 minute cap, with ±20% jitter so bursts don't line up. After 8 attempts a delivery moves to the dead-letter queue, where

Tasks 1/3

Read the retry worker
Writing docs/webhooks.md
Link it from the README

Queue another message…

PlanOpusWork

Every account, side by side

Keep work, personal and client accounts apart, with each chat showing which one it's on and what it has used.

Crashes don't cost you the chat

Chats run in a background host. If Claude Code or the app restarts, every chat comes back where it was, and duplicates are closed for you.

A real browser, every login

Chromium tabs with a profile per account, signed in at once, side by side. Chrome extensions included.

Your day, in the loop

Inbox, calendar and routines on the rails, so the meeting in 25 minutes doesn't sneak up on you mid-refactor.

Put the heavy work on another machine Coming soon

Run chats and terminals on other machines on your network over SSH, so builds and test suites don't slow your laptop. No vendor cloud in between.

Knowledge

Everything you know,
one keystroke away.

Files, repos, notes, chats and routines, drawn as a map you can fly through. Press⌘K and the answer is there, with the file it came from.

Laika Orbit's knowledge map: 745 indexed files arranged in six coloured arms (documents, notes, chats, repos, skills and routines) around a tilted ring. The spotlight is open with the query "how do we roll back a deploy". Its top answer is notes/deploy.md, section Rolling back: "Run make rollback TAG=previous and watch the canary for ten minutes", lines 3 to 6, found in 0.9 milliseconds with 0 model tokens. On the map that note glows orange, with curved connectors to five related files.

Laika Orbit recall

Context without
the token bill.

Agents answer questions about your work by grepping and reading whole files. Laika Orbit recall finds the exact section in plain code, then hands Claude one small answer. It's the engine inside the app, and it works in any Claude Code session today.

~1ms: to answer
0: model calls to search
97.8%: fewer tokens than grep-and-read¹

How recall works →

zsh

$ claude mcp add laikaorbit -- npx -y laikaorbit mcp
Added stdio MCP server laikaorbit

$ npx laikaorbit recall "how do I roll back a deploy"
tokens: roll, back, deploy · margin 82%
   11  100%  notes/deploy.md
    2   18%  brain/routers/OPS.md

notes/deploy.md › Rolling back (lines 3-6)
  Run `make rollback TAG=<previous>` and watch the
  canary for ten minutes.

151B read · 0 hop · total 0.9ms · 0 tokens

Compare

Not another editor.
The cockpit around your agents.

AI editors and agent apps are great at one repo and one conversation. Orbit is built for everything around them. Use it alongside the editor you already love.

	AI code editorsVS Code, Cursor	Agent appsClaude Code desktop, Codex	Laika Orbit
Several agents running side by side	●	●	●
A browser with a profile per account, and extensions	◐	◐	●
Inbox and calendar in the same window	—	—	●
A map of your files, repos and chats	◐	◐	●
Retrieval with zero model calls	—	—	●
Several Claude accounts at once, with usage	◐	—	●
Runs on your machines, not a vendor cloud	◐	◐	●
Open source	◐	—	●

● yes · ◐ partly · — no. From each product's public docs, September 2026. Laika Orbit runs Claude Code and is not affiliated with Anthropic.

Open source

Yours to read, run and change.

MIT licensed. Local-first: the server listens only on your machine, and nothing is sent to us. Bring your own Claude account.

View on GitHub Privacy and security

Ready for liftoff?

Clone it, run it, and put your whole working day in one orbit.

Get started for free Read the docs

Mission controlfor your AI agents

Your editor runs an agent.Your day runs on twelve windows.

Many agents. One view.

Every account, side by side

Crashes don't cost you the chat

A real browser, every login

Your day, in the loop

Put the heavy work on another machine Coming soon

Everything you know,one keystroke away.

Context withoutthe token bill.

Not another editor.The cockpit around your agents.

Yours to read, run and change.

Ready for liftoff?

Mission control
for your AI agents

Your editor runs an agent.
Your day runs on twelve windows.

Everything you know,
one keystroke away.

Context without
the token bill.

Not another editor.
The cockpit around your agents.