An agentic code editor by Ephemerent

Parallel agents,
in orbit.

Orrery is a code editor you talk to. Give it a goal and it splits the work across several AI agents running at once — each in its own copy of your repo — checks their results, keeps the best, and merges them back. Pick the models you want, watch every step, and a 3D buddy talks you through it as it works.

works with Ollama · localOpenAIGeminiDeepSeekAnthropic

Ollama, OpenAI, Gemini, DeepSeek, Anthropic — mix providers task by task.

scroll ↓
01describetell it the goal
02it editsacross many files
03it testsruns your suite
04it fixescorrects what breaks
Plan or Act mode · you approve every change · roll back to any checkpoint
Parallel agents

Run many agents at once — and keep the best.

Orrery breaks your goal into separate tasks and gives each one its own agent, working in its own copy of the repo so they never collide. Each task is attempted several times; an automatic reviewer scores the attempts and keeps the winner before merging it back. And when a job doesn't need parallel agents, Orrery won't spin them up.

01PlanSplit the goal into tasks.
02IsolateOne repo copy per task.
03TrySeveral attempts each.
04JudgeScore and pick a winner.
05MergeCombine the winners.
06MeasureProve it beat one agent.
Your buddy

A buddy that watches — and reacts.

A friendly 3D character sits beside your work, run by its own separate model. It watches what the coding agent does and reacts in real time — focused while it thinks, worried when tests fail, pleased when they pass. It talks out loud and listens back, and it stays completely out of the coding agent's way.

The tutor dial

Set how much it explains

walk me through itjust do it
Patient tutor. It narrates each step and waits for your nod; the buddy explains the why, gently.
3D review

Review for work you can actually see.

Some work isn't text — a shader, a 3D part, a game scene. Orrery can build these, try several versions, and pick the best one by a real test: does it compile, does it hit the target, does it pass an automated playtest. Then it shows you a clear before-and-after. Drag the slider to compare — no graphics card required.

ShadersWrite it, check it compiles, score the look.
3D partsInspect, compare, check it's printable.
Game scenesCollision and an automated playtest.
SplatsPreview and inspect point-cloud captures.
Model-agnostic

Local is free.
Cloud is Pro.

Point at Ollama for free local inference. Pro adds hosted cloud model access through the Orrery account, with included API credits and one interface for the whole swarm.

Ollamalocalno key required
OpenAIPro cloudincluded credits
GeminiPro cloudincluded credits
DeepSeekPro cloudincluded credits
AnthropicPro cloudincluded credits
Orrery
open project
model ollama · qwen2.5-coder (local)
# ready
 
“add a splat viewer and make the tests pass”
plan 4 tasks · running in parallel
verify tests typecheck lint
buddy “green across the board — shipping it”
Built for trust

Everything it does stays in view.

Orrery keeps a running log of every decision, test result and review as it works — all in one place. You can see why it chose an approach, what each agent tried, and what it cost, instead of digging through hidden traces.

Checked before it ships

Every change is run past tests, type-checking, linting, security and code review before it's kept.

Knows your codebase

Ask about the whole project — it searches your code by meaning, not just keywords, with @codebase.

Remembers across sessions

It recalls past decisions and your preferences from one session to the next.

Talk and interrupt

Speak to it naturally and cut in any time — it stops to listen, then picks back up.

Connects to your tools

Plugs into external tools and data through MCP, with code-aware search built in.

Measure it over time

Write your own checks for the agent and track how it performs as the models change.

Works with other editors

Plugs into other editors and agents through open, standard protocols.

Set the pace

Dial how much it explains — from quiet autopilot to step-by-step teaching.

Pricing

Local is free. Cloud is Pro, Max, or Ultra.

Orrery stays local-first: the desktop editor, local models, local logs, and core workflow are free. Every paid tier includes cloud sign-in, the BuddyIDE Pro features, bring-your-own-key, and two bundled models with monthly token quotas — pick the quota size that fits.

Free $0 forever

Everything you need to run Orrery on your own machine.

  • All local editor features
  • Ollama and other local models
  • Local audit logs and checkpoints
Get Free
Pro $40 per month

Every cloud capability for heavier, guided work.

  • DeepSeek Flash — 100M tokens/mo
  • DeepSeek V4 Pro — 15M tokens/mo
  • Bring your own API key (Anthropic, OpenAI, …)
  • Buddy narrator + cloud features
Start Pro
Max $100 per month

Bigger bundled quotas for daily swarm work.

  • DeepSeek Flash — 400M tokens/mo
  • DeepSeek V4 Pro — 50M tokens/mo
  • Bring your own API key (Anthropic, OpenAI, …)
  • Buddy narrator + cloud features
Start Max
Ultra $200 per month

The biggest quotas, for teams and heavy automation.

  • DeepSeek Flash — 1B tokens/mo
  • DeepSeek V4 Pro — 150M tokens/mo
  • Bring your own API key (Anthropic, OpenAI, …)
  • Buddy narrator + cloud features
Start Ultra

Paid tiers are monthly subscriptions billed in advance via Stripe and renew automatically until canceled. Token quotas reset each billing month and don’t roll over. Cancel anytime — access runs through the end of your paid period. See Terms and Privacy.

How it's built

A native editor and an agent core.

Under the hood, a fast native editor pairs with a separate agent engine, joined by one stable contract. Each part is built to run on its own — offline, predictable and testable.

The editor · native
Fast bufferHandles huge files without lag.
Syntax & multi-fileAccurate highlighting; edit many files as one.
3D workspaceA native model for scenes and parts.
Render proofVerifies visuals, with or without a GPU.
Buddy brainDrives the 3D buddy inside the app.
SessionsKeeps your place and history.
The agent core
The loopPlan, edit, test, fix — with your approval.
OrchestratorSplits work, runs agents, merges winners.
3D layerBuilds and checks shaders, parts and scenes.
The buddyA separate model that narrates the work.
Model providersLocal free, Pro cloud, behind one interface.
Search & voiceCodebase search, memory and live voice.
Ways to run it
Desktop appThe native editor with the live buddy.
Browser clientChat, change review and the 3D buddy.
Other editorsConnect through standard protocols.
PR reviewReviews pull requests and fixes CI.
Eval harnessMeasures the agent as models change.
Guided demoA narrated walk-through of the stack.
Bring your own models

Bring the whole swarm
down to earth.

Get Orrery, pick a model, and give it your first goal.