For developers · Claude Code & Codex

Your agent knows the code.
Now it knows why.

Decisions get made in standups, Slack threads, PR reviews and Notion specs, everywhere except the repo. KIAgent keeps all of it on your own machine, next to your Claude Code and Codex sessions, and hands it to your agent over MCP. Your agent starts from what the team agreed, you stop re-explaining, and your standup update is written before you sit down.

Claude Code checks the meeting before it touches the code.
The context gap

The code is in the repo. The reasons aren’t.

Why three retries and not five? Who said the breaker was out of scope? It was decided out loud on Tuesday, half-confirmed in Slack, and lives on in an agent session you closed on Thursday. Your agent sees none of it, so you paste it in again, or it guesses.

Pick up where you left off

Back from the weekend, already caught up.

Back from a weekend or a week off? Ask Claude what happened while you were gone. It reads the Slack threads, PR reviews, spec edits and mail you missed, and tells you what shipped, what broke, what was handed to you and who’s waiting. It also checks your own agent sessions, so it knows where you left off.

  • Ask in Claude Code, Codex or any MCP client, with no new app to open
  • Updates from everyone: Slack, GitHub reviews, Notion edits and mail
  • Your own Claude Code and Codex sessions, down to the todos still open
  • Every line comes with a source you can open
Monday morning: “what happened while I was gone?”
The status update

Your standup, written before you sit down.

Standups and status updates ask you to remember what you did. KIAgent already saw it: commits and PRs as they land, the sessions your agents ran, the threads you answered. A few minutes before your daily, a brief opens with yesterday, today and what’s blocking you, built from what actually happened and not from what the board says.

  • Your daily is a meeting, so its brief is ready a few minutes before
  • Built from commits, PRs, agent sessions and meetings, even ones nobody moved on the board
  • Edit the instructions to match your team’s format
  • Ask your AI to draft it for #standup. It waits in your Outbox until you press Send
The standup brief: yesterday, today, blocked, with a source on every line.
Sources

Everywhere the reasons live.

Code & agents
G
GitHub
C
Claude Code
X
Codex
Meetings
●
Meeting recorder
31
Google Calendar
O
Outlook calendar
Team talk
Slack
Gmail
Microsoft 365
Specs & docs
N
Notion
D
Google Docs
Local folders

Jira and Linear aren’t here yet · request them on GitHub

Built for work machines

Context in. Your code stays put.

KIAgent is a read-only memory of the work around your code. When your agent asks it something, only the answer goes to that agent. Everything else stays on your machine, and every lookup is on the record.

It doesn’t read your code
From GitHub it keeps commit messages, pull requests, issues, reviews and boards. No diffs, no file contents.
Agent sessions, read-only
Claude Code and Codex history is read from their folders in your home directory. The consent screen lists exactly which.
It stays on your laptop
The index lives on your machine. Nothing is sent to our servers, so there’s no new vendor holding your team’s data.
Local model, if you want it
Transcripts, briefs and summaries can all be written on this machine.
Every agent lookup is logged
See which tool asked what, and when.
Open source, MIT
Read every line, or build it yourself.
Your AI asked — every request, every app, on your machine.
Honest answers

The questions a skeptic actually asks.

No. From GitHub it keeps commit messages, pull requests, issues, reviews and project boards, with no diffs and no file contents. Claude Code and Codex sessions are kept as conversations, plans and tasks.

Anything that speaks MCP: Claude Code, Codex, Claude Desktop, VS Code and other MCP clients. On your own machine there’s no account and no key to set up.

Not yet. Today the tracker side comes from GitHub issues, pull requests and Projects. If you need one of them, open a request on GitHub. It helps us decide what to build next.

It runs entirely on your machine and sends nothing to our servers. Whether your company is fine with indexing work Slack and mail locally is its call. KIAgent shows every source and every lookup, so you can answer that question honestly.

Sources keep syncing in the background while the app runs. New commits, pull requests, messages and agent sessions show up on the next sync, without anyone updating a ticket.

Your daily standup is a meeting on your calendar, so KIAgent writes its brief a few minutes before, like every other meeting. You can edit the brief’s instructions to match your team’s format. It’s written on your ChatGPT plan or a local model.

Yes. Everything on your own machine is free and open source. The only paid option is remote access, so your phone or the web can reach your machine.

Pricing

Free is the whole product.

The app does everything on your own machine, for nothing. The only paid option exists for one reason: reaching your machine from somewhere else.

Optional

Remote $5/mo FREE

a secure address for your machine
  • Reach your machine from your phone
  • ChatGPT and Claude on the web, from anywhere
  • A personal address like you.localkiagent.com
  • Cancel anytime · your local app stays free
Try for free
KIAgent

Give your agent the whole story.

Download — freeMac & Windows

The repo says what changed. KIAgent remembers why.