EtherDocs
Guide/Introduction

What is Ether?

Ether is an operating system for LLM agents — infrastructure that carries work forward on your behalf instead of trapping everything in a single chat window.

If you have used a coding assistant in an IDE, you know the pattern: you type, the model edits files, you watch the diff. Ether inverts the split. Ether owns the task lifecycle (create → classify → track → deliver). Your software owns execution (models, tools, repos, local machines). Both sides talk through documented HTTP and WebSocket contracts.

What problem Ether solves

Most “AI products” are really chat apps with extras. They forget context between sessions, lose track of multi-step work, and leave integrators to reinvent task IDs, progress, and “done means done.”

Ether treats work like an OS treats processes:

  1. Something is requested — “Fix the checkout bug,” “Summarize this quarter,” “Email the vendor.”
  2. The system interprets it — domain, pipeline, definition of done, delivery shape.
  3. Execution runs — in-process agents and/or external harnesses you provide.
  4. The user gets closure — progress while it runs; a structured result when it finishes; pause when input is needed.

You stay in the loop for judgment and closure, not step-by-step babysitting.

Ether vs a chatbot vs an IDE agent

ChatbotIDE agentEther
Unit of workMessage threadOpen folder + chatTask with ID + delivery contract
Who orchestratesUser in chatUser in IDEEther OS
Who executesSame appSame appHarness (yours or Ether’s)
Memory & deliveryAd hocLocal / product-specificPlatform — corpus, timeline, delivery
Integration surfaceOften noneExtension APIsHarness API + SDK

Core concepts

Thirty seconds of vocabulary you will see throughout these docs.

  • Task — A unit of work created from a user prompt. Has status, progress, timeline, and eventually a delivery payload.
  • Ether OS / ether-api — The backend at api.tryether.ai. Auth, tasks, context, harness leases, realtime, delivery.
  • Harness — Any external app that claims a task, runs the model/tool loop, and submits a result. Ether Code is the reference harness.
  • Execution bundle — What Ether hands the harness after claim: prompt, repo scope, context doc IDs, callback URLs.
  • Delivery — When a task succeeds, Ether composes a task.delivered payload (blocks your UI can render) and marks the task complete.

What Ether is not

  • Not a model provider — Bring OpenAI, Anthropic, or your own models in your harness.
  • Not a Git host — Ether scopes work to repos/branches; your harness uses git/GitHub.
  • Not a replacement for your IDE — Ether Code (or your agent) can run locally; Ether does not need to be the editor.