Medium
Run tests
npm test -- auth
In the private alphaAgents with seatbelts.
Olli is the safe development IDE. Its agent writes and runs code inside a sandbox, asks before anything risky, shows the cost before you press Enter, and learns how your team works.
For macOS, Windows and Linux. Works with the models you already use.
Olli · Planning
You: Add rate limiting to the login route.
Plan
- Add a limiter to src/auth/login.ts
- Add tests for the sixth attempt
- Run the auth tests
Estimate $0.20 to $0.45Low confidence: first run in this repo
PlanningSandbox onmodel pinned$0.20 / $5.00Safe by default
Watch it refuse rm -rf ~
Every action the agent wants to take goes through one checkpoint: your policy, your approval if needed, a snapshot, then the sandbox. If a rule says no, Olli tells you which rule and why.
In the private alpha
rm -rf ~BlockedOutside your workspace. Always denied.git push --force origin mainBlockedForce-push to a protected branch.terraform destroyAsks youDestructive and hard to undo. Needs your approval.npm testAllowedReads your project and runs in the sandbox.
Predictable cost
Know the cost before you press Enter
No surprise bills. You see a range before, a hard cap during and a receipt after.
Before: an estimate
A cost range and the model next to the Run button, before anything is spent. New repos say "low confidence" instead of guessing.
During: a hard cap
Session, daily and monthly budgets. Olli checks the worst case before every model call and stops at the cap. It never raises its own limit.
After: a receipt
Every model call, token and tool, estimate against actual, and any model change with its reason. Export it as JSON or CSV.
Stopped at your $5.00 session cap.
$5.00 / $5.00
In the private alpha
Checkpoints in this session
Before: edit src/auth/login.ts10:42 AM
Before: add tests/login.spec.ts10:43 AM
Before: run npm install10:45 AM
Before: edit package.json10:46 AM
Fearless
Undo anything, one step or all
Olli takes a snapshot before every change it makes, in a separate history that never touches your own git. Go back to before any action, any run or the whole session. Even a restore can be undone.
In the private alpha
Private
Your code stays on your machine
Olli is built local-first. The agent works where your code already is.
Local first
Olli runs on your machine. Your source code does not leave it unless you turn on a feature that needs it, and every such feature is listed on one settings page.
Your keys, your provider
Bring your own model keys. They live in your operating system keychain and calls go straight to the provider you chose.
No watching
Telemetry is off by default. Secrets are added at the network proxy, so they never exist inside the sandbox where the agent works.
Skills that compound
Your team's know-how, eval-proven
The agent your whole company trains. Good work becomes a skill, skills are tested before they spread, and they keep improving only when the evidence says so.
In the private alpha
Save a good session as a skill
When Olli does something well, keep it. Skills use the open SKILL.md format, so they still work if you ever leave.
In the private alpha
Reads the rules you already have
AGENTS.md, CLAUDE.md, .cursor/rules and friends are picked up as they are. Import your setup from other tools.
Coming in beta
Share across the team
A private registry for your organisation, with versions, pinning, rollback and approvals.
Coming in beta
Prove it with evals
Test a skill with and without it before rollout. Nothing is promoted on vibes.
Coming at launch
Gets better every week
A learning loop that improves skills only when the evals say the change is better.
Browse skills and Agent Packs
See what is coming to the marketplace.
Work from anywhere
Approve from your phone. Keep your code on your machine.
Start a task, walk away, and approve the next step from your phone. The phone sees an end-to-end encrypted card, never your repository.
Coming in beta
Waiting for you
The marketplace
Hire an agent with a verified track record
Agents and skills, certified by Oltin and ranked on what they actually did, not on what they claim. They run only in your sandbox, can only open pull requests, and never see more than your policy allows.
Explore the marketplaceComing at launch
@oltin/flaky-test-doctor
Certified- Success rate
- Insufficient data
- Median cost
- Insufficient data
- Incidents
- Insufficient data
- Revert rate
- Insufficient data
Measured by Oltin. Never self-reported. Numbers appear only after 30 measured jobs.
Meet Olli
You always know what it's doing
Olli shows its state with a face and a word. One glance at the status bar tells you whether it is working, waiting for you or done.
Ready
Waiting for a task.
Working
Planning, running or checking its work.
Waiting for you
Needs an approval or a plan review.
Stopped
Hit a budget, a loop or a pause.
Done
Finished, with a receipt.
Blocked
Denied by policy or failed safely.
Honest roadmap
What's in Olli today, and what's next
We label every feature with where it really is. Dates are plans, and we only call something available once it ships.
In the private alpha
Private alpha
- Olli IDE with the agent built in
- Sandbox on macOS, Windows and Linux
- Policy engine and command firewall
- Checkpoints and undo
- Estimates, budgets and receipts
- Any model with your own keys
- Save as skill
Coming in beta
Beta
- Approve from your phone
- Mission Control for all sessions
- Team skills registry and evals
- Risk-tiered approvals
- First industry Agent Packs
- Dependency and code security scans
Coming at launch
Launch
- Skills and Agent Marketplace
- Certified agents with scorecards
- Run Claude Code, Codex and Gemini CLI inside Olli
- Hand off to the cloud when your laptop sleeps
- Security agent that proves before it reports
Built for developers and teams who already use AI agents and want them safe, affordable and smarter every week.
Questions
Anything else? Write to support@oltinolli.com.
What is Olli?
Olli is a development IDE with an AI agent built in, made by Oltin. The agent plans, edits and runs code for you, inside a sandbox, with a budget, and with your approval for anything risky.
Which models does it work with?
Anthropic, OpenAI and Google models, and OpenAI-compatible endpoints such as Ollama and LM Studio. You bring your own keys, and Olli never swaps your model without telling you.
Does my code leave my machine?
Not by default. Olli runs locally, and model calls go straight from your machine to the provider you chose. Features that need to send something elsewhere are off until you turn them on, and they are all listed in one place.
What can the agent do without asking?
That is up to your policy. Safe, read-only work runs on its own. Risky actions ask you first. Some actions, like deleting files outside your project, are always blocked and no setting can allow them.
How do I get Olli?
Olli is in a private alpha. Request access and we will send an invite with the installer for your system. Public downloads open at launch.
Is there a marketplace?
Yes, it is coming at launch: skills, Agent Packs for industries such as fintech and healthcare, and agents you can hire, each certified and measured by Oltin.
How do I reach a person?
Write to support@oltinolli.com. Sales questions, support and security reports all go there, and a real person replies.
Be one of the first teams on Olli
Olli is in a private alpha with a small group of developers. Tell us what you build and we'll be in touch.