Software that
proves its work.
Aion runs on loop engineering — plan, code, test, ship, monitor, sharper every pass. Community runs the whole loop on your machine, under your gates, free forever. The enterprise pilot points the same engine at approved security findings: one bounded task, your runner, a draft PR with evidence — never a merge or deploy.
Zero spam — one note the day your seat is ready.
MITOpen source under the MIT license — the full loop is free forever on your machine. Running security remediation at work? See the governed pilot →
Two ways to run it
Same engine. Two boundaries.
The loop is the product: fresh-context phases, durable checkpoints, objective checks, bounded retries, human gates. Community gives you all of it locally. The enterprise pilot wraps it in a governance boundary for security findings — and deliberately stops at the pull request.
Community · available now
The whole loop, on your machine
Community runs the full plan → code → test → ship → monitor loop locally: fresh-context phases, durable checkpoints, human merge and deploy gates, and a local cycle trail. MIT-licensed, free forever.
- ✓Missions in, verified PRs out — failures loop back, never forward
- ✓Merge and deploy wait at your gates; you can automate them locally when you trust the loop
- ✓Every integration and skill included; Claude by default, Codex or custom providers if you prefer
Enterprise · design-partner pilot
Governed remediation, bounded by contract
The enterprise design-partner pilot ingests an approved SARIF or CodeQL finding, creates one bounded authorization, runs in a customer-controlled disposable runner, tests and rescans the exact commit, and opens a draft pull request with verifiable evidence. It never directly merges or deploys.
- ✓One approved finding → one human-authorized, bounded task — no arbitrary prompts
- ✓Runs in a disposable runner you operate; no standing vendor access to your environment
- ✓Ends at a draft PR plus commit-bound evidence — merge and deploy stay yours, always
Every capability on this site carries an explicit status — see the capability matrix.
What is Aion · Community edition
An engineering agent that owns the whole loop.
The Community edition runs on your machine. Work arrives as missions; each mission executes as a cycle through the full plan → code → test → ship → monitor loop. Production itself files the next mission. You give intent and sit at the gates — Aion does the rest, and gets sharper every pass. Five nouns are all you need to keep straight.
Mission
One sentence of intent — “export reports as CSV”, or a live Sentry error. Ranked by severity, deduped by fingerprint.
Cycle
One full pass of the loop executing a mission. Checkpointed to disk after every step — kill it mid-run and it resumes.
Monitor
Polls Sentry and PostHog and turns real incidents into missions automatically. Production files the next job.
Playbook
Your per-repo contract: how to test, build, run, deploy — and which steps wait for a human.
Brain
The agent that thinks (Claude by default). Each phase is a fresh process: no chat history, no hidden state.
The loops
It doesn’t stop at the merge.
Shipping a feature is one loop. Aion runs the loops that keep a product winning after launch — search, product analytics, reliability, performance — each one reading real production data and filing its own next move. Every one of them is the Community edition working on aionagent.app itself, under its owner's gates.
Keeps SEO at the highest level
Rankings that climb while you sleep.
Search signals rot quietly — a stale meta tag here, a broken bit of schema there, a Core Web Vital slipping after a deploy. By the time rankings dip, the cause is three releases back.
- 1
Crawls every route and audits titles, meta, JSON-LD, sitemaps and Core Web Vitals.
- 2
Rewrites the weak signals and proves the Lighthouse score went up, never down.
- 3
Opens a PR, ships it, then re-crawls to confirm the organic result actually moved.
#1 organic · 100 Lighthouse — re-checked on every deploy.
Develops UX from PostHog data
The funnel tells it where to dig.
Your analytics already know exactly where people quit. But reading the funnel, forming a hypothesis, building the variant and measuring the lift is a week of attention that never frees up.
- 1
Reads PostHog funnels and session data to find the step that bleeds users.
- 2
Forms a hypothesis, ships a variant behind a flag, and instruments the experiment.
- 3
Keeps the winning cohort, reverts the loser, and files the next drop-off as a mission.
Activation +67% · conversion 2.0% → 3.9% — measured, not guessed.
Fixes errors from Sentry data
Incidents close themselves.
Errors arrive faster than anyone can triage them. The fix is usually small, but nobody has the context at 2am — so the alert just keeps firing into an empty channel.
- 1
Polls Sentry, dedupes by fingerprint, and ranks issues by blast radius.
- 2
Reproduces the bug, traces the root cause, writes the fix and the missing test.
- 3
Ships behind your merge gate and watches the event rate fall back to zero.
Root-caused · shipped · 0 unresolved — without a ticket filed.
Holds Core Web Vitals in the green
Fast stays fast, deploy after deploy.
Speed regresses one innocent import at a time. A heavy dependency lands, LCP creeps up, a layout shift sneaks in — and nobody notices until rankings and conversion quietly follow it down.
- 1
Measures LCP, INP and CLS on real routes and diffs them against the budget.
- 2
Traces a regression to the commit that caused it and trims the weight back out.
- 3
Blocks any PR that would bust the JS budget, then re-measures to prove the win.
All vitals green · 100 performance · budget enforced on every PR.
Proof, not promises — every loop above is Community Aion running on aionagent.app itself.
0
loops keep this site sharp after every launch
0%
of merges auto-deploy to production
0
analytics trackers until you opt in
green
Core Web Vitals — budget enforced on every PR
The app
Watch production work itself off.
Every signal becomes a ranked mission. Every mission runs as an auditable cycle. You approve at the gates — one click — and inspect a branch, a PR, a deploy id, and a verdict for each one.
Mission queue
4 active · 1 awaiting youTypeError: cannot read length of undefined
cleanshot-demo
trail branch=fix/thumb-svg · pr=#214 · verify ✓ health 200
Users can export reports as CSV
reports-api
PR opened · tests ✓ · awaiting merge approval
ApproveTrend collapse on /signup conversion
web
Flaky test in checkout suite
web
Everything an engineer does — end to end.
Delivers working code
Give it an outcome instead of instructions. It designs the change, writes it, proves it with tests, and opens the PR.
Lives in your repo
Branches, commits, pull requests, merges — all inside your repository, matching the conventions you already use.
Heals production
It tracks Sentry and PostHog, traces a failure to its root cause, and ships the patch — no ticket filed, no one paged.
Teaches itself
It writes new skills, rewrites its own workflow, and banks what it learns each pass — an agent that keeps upgrading the agent.
and it compounds with every pass
The whole toolkit, already inside.
0
skills on tap, from GitHub workflows to wrangling PDFs
0+
tools it reaches for unprompted
0
toolsets it composes per mission
Install · MIT open source
Run it in one line.
Aion is open source and runs on your machine — no account, no card. Paste one command, or grab a prebuilt build from GitHub. You hold the keys and the audit trail.
One command, or a prebuilt build. macOS & Linux, arm64 and x86-64.
Drop in your key — or run Community, free forever.
One sentence of intent. It plans, ships, and verifies.
Solutions
One loop, pointed at your hardest work.
The same recursive cycle, applied to the problems that actually eat your week.
Keep production green
Incidents fix themselves before you read the alert.
Learn more →Burn down the backlog
Ship features from a sentence of intent.
Learn more →Incident → PR, autonomously
A production error becomes a merged pull request.
Learn more →Migrations & sweeps at scale
Mechanical change across the whole repo, verified each time.
Learn more →Security hardening on autopilot
DDoS defense, vuln scanning, secrets audits — preinstalled.
Learn more →Stay shipping while you sleep
A queue that works itself, overnight and on.
Learn more →FAQ
Questions, answered.
The short version of what Aion is, where it runs, and how to start.
What is Aion?
Aion is governed remediation infrastructure built on loop engineering. The open-source Community edition runs the full plan → code → test → ship → monitor loop on your own machine, under your gates, free forever. The enterprise design-partner pilot points the same engine at approved CodeQL/SARIF security findings: one bounded task, executed in a runner you control, ending at a draft pull request with verifiable evidence — never a merge or deploy.
What is loop engineering?
Loop engineering is running software work as a state machine instead of a chat: plan → code → test → build → verify → ship → monitor, with fresh-context phases, durable checkpoints, bounded retries, and human gates. Failures loop back, never forward — a fix is never shipped unverified. That engine is what both the Community edition and the enterprise pilot run on.
What does the enterprise pilot actually do?
It ingests a security finding your scanner already approved (CodeQL or any SARIF 2.1.0 source), creates one bounded, human-authorized task, executes it in a customer-controlled disposable runner, re-tests and re-scans the exact commit, and opens a draft pull request together with commit-bound evidence. It never merges, never deploys, and has no standing access to your environment. It is a design-partner pilot today, not a generally available product — capability statuses are on the capability matrix page.
How is Aion different from Copilot or Cursor?
Copilot and Cursor autocomplete while you drive. Community Aion owns the whole loop end to end on your machine: it plans, codes, tests, opens the PR, and after the merge keeps working — SEO, UX from PostHog, errors from Sentry, Core Web Vitals. You sit at the approval gates instead of writing every step. The enterprise pilot is narrower on purpose: one approved finding in, one draft PR plus evidence out.
Does Aion merge or deploy on its own?
In Community, merge and deploy default to your human gates; a local owner can enable supported local automation once they trust the loop. In the enterprise pilot, never: the workflow terminates at a draft pull request plus evidence, and there is no setting that enables an automatic merge or deploy.
Where does my code go — is it safe?
Community runs locally inside repositories you explicitly register: branches, commits and pull requests through your own credentials, with prompts going only to the model you choose. In the enterprise pilot design, execution happens in a disposable runner inside the customer environment, with controlled egress and no standing vendor access — source code and durable credentials never flow to the vendor.
Which AI model does Aion use?
Claude is the default reasoning engine; Codex is supported, and custom providers can be configured — capabilities differ by provider. Each phase of a cycle runs as a fresh process with no hidden chat history or carried-over state, so every step is auditable and reproducible.
Is there a free version, and how do I start?
Yes — Community is free forever and MIT-licensed. Install on macOS or Linux with a single command (curl -fsSL https://aionagent.app/install.sh | sh), run aion login, then give it a mission. Pro and Team add support and the current Team service. The enterprise remediation pilot is by application only.
scope → ship → watch → improve ↻
Give it the mission.
Install Community and put the loop on your backlog — or apply for the governed remediation pilot at work.