Open source · Shoofly Governance In the Loop (GITL)

A human in the loop for the AI your team already uses

AI governance and human-in-the-loop approval in one open source system. Your team works in Claude chat, the CLI or Slack. Every action is governed, filtered and logged, and a person approves what matters.

Claude chat, CLI and Slack todayChatGPT and Microsoft Teams soon

The problem

Your AI lives in two places. Your data should live in one.

Company AI lives in Slack

The teammate that knows your clients, deals and calendar works with the right permissions.

People think in a chat app

Your team drafts, plans and reasons in Claude or ChatGPT all day. That is where the work actually starts.

Copy-paste is the leak

Moving between the two means pasting client detail into chats no one governs: no approvals, no filter, no audit trail.

GITL closes the gap: the chat app gets your teammate's answer, never your systems.

What it is

One connector. Three things your team can do from the chat.

Talk to your crew member

Ask your own AI teammate and get its reply. It answers with your own permissions.

See and file requests

Your submissions, access and tool requests, and what is waiting for a decision. Filing only puts it in the queue.

Decide approvals

Approvers approve, send back or decline. The chat app asks them to confirm every decision, and it is recorded.

Plus: teach your crew member what you work out in the chat. More on that in a minute.

How it works

The chat app asks. Your stack answers. Nothing else crosses.

Setup, about three minutes

Add it once. Sign in with your company account.

Add the connector

In Claude.ai: Settings, Connectors, Add custom connector. Paste the Shoofly address and keep the detected sign-in options.

Sign in and approve

A Shoofly page shows which app is asking and exactly what it may do. Approve, and you land back in the chat, connected.

No keys, no passwords to copy. The chat app never sees one.

Claude.ai add custom connector screen with sign-in options detected

Claude.ai, add connector

Shoofly approval page

Screens: company view, Oct 2026.

Setup, step 3

Claude.ai tool permissions for the Shoofly connector, every tool set to needs approval

Claude.ai tool permissions, company view, Oct 2026

You choose what runs without asking.

Every tool starts at Needs approval.
Read-only tools such as Who am I and List the queue can be set to Always allow.
Anything that writes, like filing a request, can stay on Needs approval.
With Needs approval, nothing leaves the chat until you click.

Governance in the loop

Approvals stay with a human. Every decision is confirmed and recorded.

STEP 1

A request is filed

From the chat or by the crew member. Connecting approves nothing; filing only queues it.

STEP 2

It waits in the queue

Visible to the person who filed it and to the people allowed to decide it.

STEP 3

A human decides

Approve, send back or decline. The chat app asks the approver to confirm each one.

STEP 4

Checked and recorded

Your company system re-checks the approver and the exact item before anything changes, then logs it.

Approvals ask twice on purpose. Approver sign-ins are re-confirmed every day, and deciding needs the company workspace.

Data stays put

The company workspace gets the work. Personal accounts get the basics.

	Company Claude workspace	Personal Claude account
Ask your crew member	Yes, with your permissions	No
Client data	Yes, minus restricted clients	Never sent
Restricted clients (regulated, BAA, NDA)	Never sent; a filter withholds any slip	Never sent
Queue, requests, teaching	Full	Status, file a request, save notes
Decide approvals	Approvers only	No
Model training on your data	Not used for training (business terms)	Client data never reaches it
What is kept	Questions and replies, 30 days	Saved notes only, in your crew member
Audit trail	Every call: who, which tool, when	Every call

The filter checks every reply before it reaches the chat app. Restricted clients, SSNs and card numbers are always withheld; on a personal account, every client detail is.

Teach your crew member

What you work out in the chat becomes what your teammate knows.

"Save this." One decision, preference or term at a time, as a clean summary.
Import your chat history. Pick the conversations; your crew member turns them into short notes.
Inbound only. Nothing comes back out except "saved", so it is safe from a personal account too.
You stay in charge. Every note is listed by title and date, and can be deleted.

Admin control

You can see it, limit it, and switch it off.

Revoke any connection

Disconnect one chat app, or every app at once when someone leaves the team.

One pause switch

Turn the connector off for everyone in one step if anything looks wrong.

Daily limits

Asks per person per day are capped, and the cap is enforced in one place.

Per-person AI budgets

Usage counts against each person's budget, visible in your company cockpit.

Sign-ins expire

Short-lived sign-ins renew automatically; approvers re-approve every day.

Counts, not conversations

Admins see usage counts and short daily summaries, never anyone's chats.

Shoofly Governance In the Loop (GITL)

See it on your team.

A short call: we show GITL running, walk through approvals, filters and the audit trail, and mark where a person should stay in the loop. Open source, or operated by Shoofly.

Book a walkthrough → Join the open source waitlist

Claude chat, CLI and Slack todayChatGPT and Teams soonshooflyai.com/gitl