I.R.I.S is an offline AI agent for macOS that can execute real workflows using your filesystem, Terminal shell, and other tools on your machine.
I.R.I.S turns natural language requests into executable plans, running them locally via Terminal, the macOS filesystem, and tools like Git—following a strict "describe, plan, approve" workflow.
User: "Update README, run tests, and push if green."
I.R.I.S Plan: "(1) edit README.md (2) run pytest (3) commit changes (4) push to origin."
I.R.I.S Request: "Approval required to push. Proceed?"
Result: "Done. Audit log saved."
Tell I.R.I.S what to do in plain English.
See the exact plan (commands & edits) before it runs.
One click to execute locally. Results logged.
Cloud agents require sending private repos, API keys, and client data to third-party servers where redaction often breaks context. Local execution on macOS eliminates this risk entirely.
Pasting credentials into any cloud service, even encrypted, exposes you to breach risk.
Code history, internal notes, and proprietary architecture shouldn't be shared with model providers.
PII, PHI, and client documents often cannot legally leave your infrastructure.
Masking sensitive data often strips the context (repo structure, configs, shell output) the AI needs to actually solve the problem.
Designed with a "paranoid-first" architecture, I.R.I.S operates offline by default, requiring explicit approvals for destructive actions and maintaining a strict audit log of every command.
No network calls required for core functionality. Your files never leave your Mac.
Dangerous steps like `git push`, file deletions, or sending data trigger a mandatory confirmation prompt.
Every shell command executed, file edited, and agent decision is logged locally for compliance.
You grant or revoke access to specific directories and tools at any time.
Built for Apple Silicon using open-weight models, I.R.I.S automates complex engineering tasks—from Git operations to ops runbooks—serving engineers, DevOps, and security researchers.
| Capability | Example Workflow |
|---|---|
| Repo Ops | Format code, run tests, commit changes, and push if successful. |
| Local Automation | Rename files in bulk, generate changelogs, update documentation. |
| Ops Runbooks | Run deployment checklists and summarize logs without sharing credentials. |
"Run tests, fix lint errors, update README, and push if green."
"Collect logs, summarize failures, and generate a status update draft."
"Audit what changed in this folder since yesterday and explain why."
Compatibility: Optimized for macOS (M1/M2/M3). Integrates with Terminal, Filesystem, AppleScript. More integrations coming.
Join the waitlist to try I.R.I.S on your Mac during the beta—free access, no credit card, and updates when new builds ship.
We'll only email product updates and beta access. Unsubscribe anytime.