Papillon Corporation
Agentic engineering
An agent pipeline your team runs.
Written requests go in, reviewed changes come out, and your people decide what ships. Set up on your own infrastructure, with hosted or local models.
01 / What a change goes through
Five steps, and a person at the end.
-
A written request
A ticket from your own queue, written by your people. No chat session.
-
An agent builds it
On your own infrastructure, with a hosted or a local model. Safety rules are enforced in code, not left to the model.
-
Tests re-run
After every step the agent takes, not once at the end.
-
The change is reviewed
It arrives from your ticket queue as a reviewed change, not a draft.
-
Your people decide
What ships is their decision, with a record of what every step cost.
02 / What you receive
Agents doing real work, not demos.
It stays inside
The pipeline runs on your own infrastructure and can use local models, so code never has to leave your network.
Rules in code
Safety rules are enforced in code, not left to the model, and tests re-run after every step an agent takes.
Every step costed
You get a record of what every step cost, so the spend on each change is a figure and not a guess.
03 / Ours, running
We build this way ourselves, every day.
The pipeline Papillon sets up for clients is the one it runs for its own products. Every figure here is a change the pipeline itself picked up, built, reviewed and merged, read from its own records.
Updated 12 Nov, 17:55
- 273
- changes merged by the pipeline since September 2026, across 4 areas of work
- 162
- in the last 30 days
- 29
- so far this week
- 46minutes
- median time from pickup to merged change
Latest deployment: Print-shop manager, version d862ee2 on 12 Nov, 17:25. Here a merged change is a deployed one.
| Area | Merged | Last 30 days | Last 12 weeks | Latest deployment |
|---|
Counts and dates only. A change is a ticket the pipeline picked up from a written request and took to a reviewed, merged pull request with no chat session. Work done any other way is not counted. Metered model spend across all of it: $33.23, or 12¢ a change. Half of the changes took longer than the median time, some of them much longer.
04 / Questions
Questions people ask first
Can AI agents work on code that must not leave our network?
Yes. The agent pipeline runs on your own infrastructure and can use local models, so nothing has to go to a hosted service.
Who decides what ships?
Your people. The pipeline takes a written request through to a reviewed change. Whether that change goes out is their decision.
How do we know what it costs to run?
Every step is recorded with its cost. Papillon's own figures are on this page: metered model spend, in total and for each change.
Want this for your team?
Say what your team builds and where its code has to stay, and get a straight answer on whether a pipeline fits.
Your details are used only to reply to you. See the privacy notice.