Orion

The AI super agent that has already read your codebase.

Orion is the AI super agent for code — it reads your whole codebase, edits across files, runs your tests, and can ship a pull request on its own. In your editor, terminal, desktop, chat tools, or phone. Cloud brain, local disk.

Download OrionGet started

Free to start · macOS, Windows & Linux

orion — agentclaude-opus

refactor auth to the new token flow and run the tests

● Read auth/session.ts · 3 usages

● Editing session.ts, middleware.ts

● Running npm test…

✓ 14 passed in 2.3s

Done — swapped the legacy token check for issueToken + verifyJwt across 2 files. Tests green.

auth/session.ts

export async function session(req) {

- const t = legacyToken(req)

+ const t = await issueToken(req)

- if (!verifyLegacy(t)) deny()

+ if (!verifyJwt(t)) deny()

return user(t)

}

Works with the models you already use

AnthropicOpenAIGrokGeminiTogether
One agent, three surfaces

Use Orion however you work

The same super agent and the same backend — in a full editor, your terminal, or a focused desktop app.

orion-ide
Orion IDE screenshot

The full editor

A full code editor with the agent built in — chat, multi-file edits, and your whole toolchain in one window.

Whole-codebase awareness

It reads your repo before it writes a line

Orion retrieves the symbols, references, and structure that actually matter for your task — so its edits fit your conventions instead of guessing at them.

  • Semantic + structural retrieval across the whole project
  • Follows imports, call sites, and types
  • Stays grounded in your real code, not a summary
orion — context

retrieving context…

◆ auth/session.ts · issueToken, verifyJwt

◆ auth/middleware.ts · requireUser

◆ lib/tokens.ts · TokenClaims

+ 4 more files · 11 references

Cloud brain, local disk

Scales like a service. Respects your machine.

The agent runs in the cloud where it can think hard and scale. Every file read, write, and command runs locally — your code and terminal never leave your machine.

  • Heavy reasoning + orchestration server-side
  • All file and shell access stays on your disk
  • Read-only, workspace-write, or full access — your call
orion — runtime
Cloud brain

Reasoning, planning, orchestration, memory.

· plan + retrieve

· call the model

· coordinate sub-agents

Your machine

Files, terminal, Git — never leave your disk.

· read / write files

· run commands

· talk to your LSP

Agentic editing

Edits across files, then proves it works

Orion plans a change, edits every file it touches, runs your tests, and self-corrects when something breaks — instead of handing you a diff and hoping.

  • Coordinated multi-file edits
  • Runs your tests and reads the output
  • Detects when it's stuck and recovers
orion — verify

applying changes · 2 files

+ const t = await issueToken(req)

- const t = legacyToken(req)

14 tests passed
Autonomous agents

Hand off the whole task, get a pull request

Point Orion at a repository and describe what you want. It works on its own — planning, editing, running your tests, and fixing what breaks — then opens a pull request for review.

  • Runs end-to-end with no turn-by-turn steering
  • Each task runs in its own isolated cloud sandbox
  • Live timeline, with per-team budgets and limits
orion — ship
feat: token-flow auth refactorOpen

orion/token-flow → main · 3 commits · +214 −96

tests 14 passed

typecheck clean

security scan no findings

Opened by orion-agent · ready for review

Built to actually ship code

Not a chatbox bolted onto an editor — a complete agentic workflow.

Whole-codebase retrieval

Reads symbols, references, and structure before it touches a line.

◆ auth/session.ts · issueToken, verifyJwt

◆ lib/tokens.ts · TokenClaims

+ 4 more files · 11 references

Agentic multi-file edits

Plans, edits across files, and validates its own changes.

+ const t = await issueToken(req)

- const t = legacyToken(req)

14 tests passed

Cloud brain, local disk

Runs in the cloud; your files and terminal never leave your machine.

Any model

Anthropic, OpenAI, Grok, Gemini and more behind one interface.

Autonomous agents

Hand it a repo and a task; it works on its own and opens a pull request.

Parallel specialists

Fans work out to planners, coders, reviewers, and debuggers at once.

Security built in

Scans every AI-generated edit and command for risk before it lands.

Sees, watches & talks

Understands screenshots, watches video you drop in, and works by voice.

Safety by default

Read-only, workspace-write, or full access — you choose.

Persistent memory

Remembers your conventions and decisions across sessions.

Runs your tools

Drives your terminal, runs your tests, talks to your language servers.

Extensible

Bring your own hooks, skills, and MCP servers.

Frequently asked questions

What exactly is Orion?

An AI super agent for code. Instead of bolting a chatbox onto an editor, the agent is the center — it reads your whole codebase, plans changes, edits across files, runs your tests, and drives your terminal. Use it in a full editor, your terminal, a desktop app, from chat tools like Slack, or on your phone — and hand it a whole task to run on its own and open a pull request.

Can Orion work on its own?

Yes. Point it at a repository and describe the task, and Orion runs end-to-end — planning, editing across files, running your tests, and fixing what breaks — then opens a pull request for review. Each autonomous run is isolated, with per-team budgets and limits so you can delegate safely.

Where does my code go?

Your files, terminal, and Git never leave your machine. Only the context the agent needs to plan its next step is sent to the cloud — file reads, writes, and commands all run locally.

Is AI-generated code safe to accept?

Orion has a security layer built in. It scans every edit and command the agent generates for risk — insecure code, leaked secrets, risky dependencies, dangerous shell commands, and prompt-injection attempts — and can surface findings or block them outright before anything lands or runs. You choose how strict it is.

How is it different from other AI editors?

It's not just an editor — it's one agent across your editor, terminal, desktop, chat tools, and phone, and it can run a whole task autonomously. Underneath: retrieval that genuinely understands your repository before it edits, and a clean split where the agent's brain runs in the cloud while everything executes on your disk.

Which models can I use?

Anthropic, OpenAI, Grok, Gemini and others — all behind one interface. Switch per session without changing how you work.

What platforms are supported?

macOS, Windows, and Linux across the Orion editor, terminal client, and desktop app — plus a control center that drives your agent from chat tools like Slack and Teams, and native iOS and Android apps for watching runs and approving actions on the go.

Is there a free plan?

Yes. The Hobby plan is free forever and lets you bring your own API key. Upgrade to Pro for included usage and frontier models.

Ship faster with an agent that knows your code

Download Orion and let it read your repo, run your tests, and do the work — while your code stays on your machine.

Download OrionSee pricing

Free to start · macOS, Windows & Linux