v0.11.20 · Public beta

Hand off the busywork. Your agents ask before they act.

Answer the email, find the time, write it up, sort the bills. Agents do it in your apps, on your own computer, and nothing goes out until you say yes.

Windows · macOS · Linux · No card · $5 to start

  • US models only
  • Anthropic
  • OpenAI
  • Google
  • xAI

Change one base URL and every call to all four providers says which model answered and what it cost.

Capabilities

Agents do the work. You make the calls. Four capabilities, all built around your yes.

Every change waits for your yes. Emails, posts, payments and file changes stop with exactly what will happen in front of you.

Approvals · PipelineNeeds you

Hand an agent a whole job. It writes a plan, you approve it, and it works in the background, then brings the result back for review.

Delegate · Repo WatcherPlan ready

Real files, not just replies. PDF and Word reports, slide decks and spreadsheets land in your Library, with a canvas beside the chat for changes.

Library · LedgerPDF · 15 KB

Memory that stays on your computer. Everything your agents remember, where each note came from, and a search across all of it.

Settings · MemoryOn this computer
Use cases

One agent per job, at home or at work

Three emails need you. All three are drafted.

Inbox drafts the replies that need you. Nothing is sent until you approve it.

Draft replies to anything that needs me3 of 31

A time that works for everyone.

Calendar reads the thread, checks your calendar and sends the invite when you say yes.

Find an evening next week for dinner with Sam and PriyaWed 7:00

Your notes, as a page to send.

Writer turns rough notes into a summary: what was decided, who does what, and by when.

Turn Tuesday’s notes into a one-page summary1 page

Three quotes, one recommendation.

Research compares the options, then recommends one. Accepting it waits for you.

Compare the three quotes for the new fence$4,860

What renews next month, and what went up.

Bills reads the receipts in your email, spots the price rise and drafts the cancellation for your yes.

What renews next month, and what went up?$39 → $49

Follow-ups that answer what was asked.

Pipeline drafts each reply from your call notes. Nothing is sent until you approve it.

Draft follow-ups for yesterday’s three demosAsks first

The overnight inbox, sorted by morning.

Help Desk drafts the routine replies and puts the urgent one in front of you.

Go through the support inbox from overnight14 emails

Reconciled, with the one thing that’s off.

Ledger checks payouts against the bank and finds the duplicate charge. The refund waits for your yes.

Reconcile September’s payouts against the bank feed$480

From release notes to a scheduled post.

Campaigns reads what shipped, writes the post and adds it to your content calendar as a draft.

Turn this week’s release notes into a launch postDraft
How it works

Lobstack connects your tools, does what you ask, and changes nothing until you say yes

The connector catalog: Notion, what it does, and the four actions an agent can take in it

Connect your tools

GitHub, Slack, Notion and Linear in one click, or any MCP server. Keys stay in your OS.

The thinking menu: Quick, Deliberate or Deep, each naming the model it uses

Ask, and choose how hard it thinks

Write it in plain language, then pick Quick, Deliberate or Deep and read its reasoning.

A run’s receipt: Repo Watcher, billed to Acme, $0.040 metered, with tokens in and out

Approve, then read the receipt

Changes wait for your yes. Every run ends with its steps, tokens and cost, billed to the right client.

News

What shipped, and what we got wrong

22 Oct 2026 · Release 0.11.18

Start from the job, set the rules

Lobstack 0.11.18: new agents start from what you want done, every agent gets its own approval rules, and every reply says which model answered and why.

Read the post
Diagram from the post: pick a job, get a ready agent, its rules checked from the top, and each reply names its model

19 Oct 2026 · Guide

How to Build a Local AI Agent Without Cloud APIs

Choose hardware, a tool-calling model, and a local runtime, then wire an agent loop with validation, approval gates, and evaluation.

Read the post
Diagram from the post: a model on your computer sends tool calls to an agent loop that checks them, and a person approves before anything changes

11 Oct 2026 · Company

Why we are Lobstack ASI

ASI is the direction AI is going. Lobstack builds the part that keeps a person in charge, and every model now comes from a US company.

Read the post
Diagram from the post: four US model providers feed Lobstack’s approval, receipt and spend cap, and a person approves
Console

One ledger for every call, every client and every cap.

50%of what Nex saves comes back to you as credit
25clients on Agency, each with a budget and a statement
3levels of spend cap: per client, person and key
50%of a person or key cap sends an email, before the hard stop
Console · Logs, Clients, Spend capsSample data
Security

It stays on your computer.

Prompts go to the model and connector traffic to its service. Everything else stays with you.

One folder on your disk

Workspace, vault, sessions and logs stay on your computer.

Keys in your OS

Encrypted, with the key in Credential Manager, Keychain or Secret Service.

No telemetry

No analytics. A report leaves only when you press Send.

Connectors & MCP

Your tools, one click. GitHub, Slack, Notion, Linear, or any MCP server.

Put the Gateway in any agent

$ npx -y @lobstack-ai/mcp

71 connectors · plus any MCP server

Changelog

The last four releases

Full changelog

Asked often,
answered plainly.

  • Free to download. A new account from the app adds $5 of credit for 30 days. After that it stays free with your own provider key or GitHub Copilot. Or use a paid plan.

  • No. It uses the Lobstack API by default, with a receipt for every call. GitHub Copilot or your own provider key also work, and neither needs a Lobstack account.

  • Your computer. Workspace, vault, sessions and logs: one folder on your disk. Prompts go to the model, connector traffic to its service. There is no telemetry or analytics. A report leaves your computer only when you press Send.

  • Encrypted, with the key in your operating system’s credential store: Credential Manager, Keychain or Secret Service. Secrets never enter the database, and values are redacted as they are written.

  • It is a public beta. The Mac app is signed and notarized by Apple and opens normally. The Windows installer is signed with a verified publisher; while the certificate is new, SmartScreen may still ask once: choose More info, then Run anyway.

  • No. The app’s repository is private. What you can check instead: the installers are signed, the security docs list every control and its state, and every call comes back with a receipt. The npm packages and the developer skill files are MIT.

  • Anyone who needs a SOC 2 report or single sign-on today: neither exists yet. Anyone who needs the app’s source, or a model from outside Anthropic, OpenAI, Google and xAI.