One approval queue
Pages, client-facing work, new agent skills and access requests land in one place. The right person decides, and an item that changed after review goes back for review.
Open source · Shoofly Governance In the Loop
Your team builds AI in Claude chat ChatGPTsoon the CLI Slack Microsoft Teamssoon GITL keeps a human in the loop
AI governance and human-in-the-loop approval in one open source system. Every person gets their own agent, every agent follows one set of rules, and anything client-facing waits for a person to approve it.
Example Approval queue
Marketing agent wants to
Publish the updated services page
Rewrote the hero and two FAQs from last week's calls.
Requested by the marketing lead
- Every decision is logged: who, what, when.
Why it exists
Your team writes the pages. GITL binds the book.
Team Builds Agent Proposes Person Approves Every step Logged
Quick answer
Shoofly GITL (Governance In the Loop) is open source software that governs the AI your team already uses. It gives each person an agent where they work, scopes what each agent can see, logs every action, caps spend, and routes anything consequential to a person for approval.
Your team is already building with AI: agents for call reviews, coaching, marketing and reporting. Each one helps. Together they run without a referee. One agent's output feeds the next, nobody checks it, and leadership cannot see what is running.
GITL is the binding. A person stamps each page before it runs company wide, every agent follows the same rules, and every page is tied to a number. One book your leadership can read.
How it works
Agents propose. People decide.
The same five steps behind every surface, every agent and every person.
- 01Someone asksFrom Claude chat, the CLI or Slack.
- 02Rules applyTheir permissions, their data, filtered.
- 03The agent proposesA page, a message, a skill, an access change.
- 04A person approvesApprove, ask for changes, or decline.
- 05Every step is loggedWho asked, what ran, who said yes.
What you get
Governance and approval, built into the work.
An agent for every person
Each teammate gets their own agent with their tools and their access level, in the places they already work. It remembers their context.
Data scoped by person
An agent works with its person's permissions. Restricted clients and sensitive fields are filtered out of chat answers. Wider access takes an approved request.
A record of every action
Who asked, which tool ran, what was approved and by whom. Admins see counts and short summaries, never anyone's chat.
Budgets that hold
Every person has an AI budget. As they near it, their agents step down to lower-cost models on their own, so the invoice holds no surprises.
Teach it from your chats
What you work out in a chat becomes what your agent knows. Saved notes are scanned first, kept as reference, and you can see and delete every one.
Where your team already works
No new app to learn.
GITL meets people in the tools they open every day, with one set of rules behind all of them.
Connect Claude to your company agent. Ask, file a request, check the queue, save what you worked out.
What is waiting on me today?
Shoofly GITL · queue status
Two items need your approval: the updated services page and a new renewal-quote skill. Both passed review.
A Claude Code plugin and a crew command bring your agent into the terminal and Cursor.
$ crew ask "which renewals close this month?"
answering with your permissions…
Three renewals close this month. Want me to draft the quotes for approval?
Each person DMs their own agent, and agents work in your channels next to the team.
You Draft the recap for this week's client call.
Your agent Drafted. It is in the approval queue because it goes to a client. Approve
Coming soon The same connector for ChatGPT, under the same rules, once it meets our data terms.
Coming soon Your agent in Teams chats and channels, under the same rules.
Two ways to run it
Open source, or operated by Shoofly.
Open source
Read every line, run it yourself and change it for your company. Built for teams that already build with AI and want governance they can inspect.
Join the waitlistGITL, operated by Shoofly
We set up GITL for your team: the agents, the rules and the approval queue. A named Shoofly person operates it, fixes what breaks and binds in what your team adds.
Book a walkthrough See it on your team
Book a GITL walkthrough.
Thirty minutes. We look at the AI your team already uses, show GITL running, and mark where a person should stay in the loop.
Widget not loading? Book on the booking page or write to us.
Open source early access
Get the source first.
The repository opens to the waitlist before it goes public. Leave your email and we will send it the day it opens.
Questions we hear
FAQ
What is Shoofly GITL?
Shoofly GITL (Governance In the Loop) is open source software that governs the AI your team already uses. Each person gets their own agent where they work, every agent follows one set of rules, every action is logged, and anything client-facing waits for a person to approve it.
What is human-in-the-loop approval?
An agent can draft and propose, and a named person decides before anything consequential happens. In GITL that means pages going live, messages to clients, new agent skills and wider data access land in one approval queue. The approver taps Approve, Ask for changes or Decline, and every decision is recorded.
What is AI governance in practice?
Knowing which agents exist, what each one can see, what they did and who approved it. GITL gives each agent its person's permissions, filters restricted data out of chat answers, caps spend per person, logs every action and gives admins one switch to pause it all.
Is GITL open source?
Yes. You can read every line, run it yourself and change it for your company. The license does not allow other firms to host or resell it for two years from each release. The public repository opens to the waitlist first.
Where does my team use it?
Where they already work: Claude chat through a connector, the command line through a Claude Code plugin (Cursor works too), and Slack, where each person has their own agent. ChatGPT and Microsoft Teams are coming soon.
Who is GITL for?
Companies whose team already builds with AI and whose leadership wants a person in the loop, visibly, before agents touch clients, money or email.
Is our data used to train models?
No. GITL works through Claude under business terms that do not train on your data. Questions and replies relayed through the connector are deleted after 30 days, and admins see counts and short summaries, never anyone's conversations.
Do we have to hire Shoofly to use it?
No. Run it yourself from the source, or have Shoofly set it up and operate it for you. Book a walkthrough and we will show you both.
If something breaks, we fix. If something gets added, we bind.
See how GITL would govern the AI your team already built, and where a person should stay in the loop.