How it starts
one API call from GitHub Actions or Azure Pipelines, a cron job or your bot
An AI agent you call from CI
An agent that files issues and drafts PRs, called from your CI.
Pintle is an API. Your pipeline sends a Slack thread, a Linear issue or a diff, and the agent, a language model reading your rules, files the issue or drafts the pull request.
Free for the first month · no card
Your prompts live in your repo as plain files. Every run is tied to a commit.
The full log of each run lands in your own storage before we call it done.
If we are bought, your prompts and logs stay where they are.
One call
from your CI, a cron job or a bot
€40
per 10,000 credits, prepaid, no organisation fee
7 days
then our copy of your run log is deleted
What you get
What the old vendor kept, and where it lives with Pintle
Stilla runs when someone mentions it, and its docs list no public API. Once it is folded into Meta’s business agent, triage and PR drafting go back to your engineers.
Prompts
plain files in your repo, changed by pull request
Run history
a full log of every run, in your storage
Credentials
your tokens travel with each request and are dropped after
Triage
your written rules first, then a search of past issues
Replay
rerun any past run against a new prompt or model
The product
What it looks like day to day
Your rules are a plain file in your repo, changed by pull request
A week of runs, and what each one did
Replay last week’s run against your branch before you merge
Your rules as a plain file, a week of results at a glance, and a replay before you merge.
How it works
What happens when you call it
Six steps, always in this order.
If your storage never confirms the log, the run fails and you aren’t charged.
Anything already filed stays filed. So do the log lines your storage accepted.
Fetch - pulls your prompt folder at the exact commit you named
Plan - a model reads your rules and decides what to do
Match - checks the obvious first: same issue key, same error, same stack trace
Score - otherwise searches your past issues and ranks the closest
Act or ask - above 0.85 confidence it files or drafts. Below that, your pipeline gets candidates and decides.
Confirm - the log reaches your storage, then you get the OK
Integrations
Four tools, all under your own app names
There is no Connect button. You register the app, create a token with the permissions listed, and send it with each run.
- Linear - files the issue, or lists the likely duplicates
- GitHub - drafts the pull request, and holds your prompts
- Slack - reads the thread you send, posts only if told to
- Notion - updates only the page you name
The matcher
Most of the machine learning answers one question: is this already filed?
A language model plans the run; for matching, a quoted issue key or identical error settles about two in five.
Search and score - embedding search compares the thread with your own Linear issues, then a classifier ranks what comes back
Below 0.85 - Pintle files nothing; the run returns ranked candidates as a proposal and stops
Your pick - a named person chooses, with a button in your Slack bot or a manual gate in CI
What it keeps - the candidate ids offered and the one chosen, never the thread text, tuning only your matching
The API
Three endpoints, plain files, and 5,000 free credits to start
Start a run, replay a run, refresh your issue index.
One run · finished
- the job
- Sort out a new incident
- your rules
- Read from your own GitHub repo, pinned to one exact version
- what came in
- A Slack thread from #incidents, 23 messages, sent by your CI
- what it found
- 91% sure this is the same problem as PLT-2291
- what it did
- Commented on PLT-2291 under your Linear app’s name, and filed nothing new
- run log
- Saved to your storage before we confirmed the run
Not a dashboard. Our console keeps keys, spend caps and seven days of runs. The full record is in your storage.
Each call says which skill to run, what to work on and which version of your prompts to use.
To start, copy our sample triage folder into your repo, open a pull request and send one request from your laptop.
You get a full log in your storage and an issue filed under your own app’s name.
Why it works this way
Changes go through pull requests, and your storage has to stay up
There is no prompt editor in our console.
The agent remembers nothing between runs. Everything it knows is in your files or the input, where you can read it.
- A product manager who wants a rule changed needs an engineer.
- Unwritten rules get written into rules.md by hand, once.
- We can’t show you last quarter. After seven days, only your storage has it.
Scope
It never starts on its own. You may miss that.
- in scope01Every run can be replayed from its log: same commit, same model, same steps.
- we don't01No chat listener. Mentioning Pintle in a Slack thread does nothing.
- in scope02The prompt folder format is open, under a permissive license.
- we don't02No one-click connect. You register your own GitHub, Linear and Slack apps and send the tokens per run.
- in scope03Twelve months’ notice before we retire an API version.
- we don't03No learning by watching, and no Gmail, Docs, calendars or meeting notes.
Pricing
€40per 10,000 credits, prepaid
One credit is 1,000 tokens, input and output together. No organisation fee, no seats. Stilla’s Team plan is $40 a month per organisation, plus credits.
- 10,000 credits€40
- 50,000 credits€175
- 250,000 credits€750
Create an account
- Sandbox key - 5,000 credits, no card. Unused credits from a first purchase are refunded within thirty days.
- Spend cap on every key - a runaway pipeline stops at the cap. Credits last twelve months.
- Never charged - tool calls, replays, index refreshes and log delivery
Run from Dublin by an Irish company that keeps seven days of your runs
- Your tokens are dropped the moment a run ends.
- Your prompts, inputs and logs are never used to train a model.
- Your issue index sits on GPU servers in Dublin, deleted thirty days after you close the account.
- The tool you pay for today states SOC 2 Type II and says nothing about where data lives or for how long.
After seven days, a run leaves nothing with us but your team’s pick when the agent wasn’t sure.
Get a sandbox key
We ask for your work email and your company’s name.
We reply within one business day.
- ✓ 5,000 credits, no card
- ✓ Three sample folders: incident triage, PR drafting, page updates
- ✓ Full runs, on the same models as a paid key
- ✓ Not for teams without an engineer to own the pipeline