# OpenAgents Docs > Understand Omega, the native OpenAgents workspace for projects, agents, code, review, and evidence. # Omega Source: https://openagents.com/docs Omega is a native, Zed-based IDE from OpenAgents. It brings project work, agents, code, review, and evidence into one workspace. Omega is in active development. Signed and notarized macOS ARM64 prerelease candidates exist for evaluation. The latest recorded candidate still has open installed-release gates. Omega is not a generally available or supported release. ## Product direction Omega starts with the editor, project, buffer, terminal, and worktree model from Zed. OpenAgents adds agent work, review, policy, evidence, and run history without creating a second project authority. The target experience keeps these parts together: - Native project and editor work - First-party and external agents - Visible plans, tools, decisions, and file changes - Code review and evidence for completed work - Bounded automation that a user can inspect and stop Not every target capability is complete in a prerelease candidate. These docs identify a direction or an open gate when current evidence does not prove availability. ## Availability There is no supported public installer in these docs. A prerelease artifact is test evidence, not a stable release. Use the [Omega repository](https://github.com/OpenAgentsInc/omega) for source, release records, open issues, and candidate notes. Read the exact candidate status before you install or evaluate an artifact. ## Documentation map - [Availability and setup](/docs/getting-started) explains the current prerelease boundary. - [Native workspace](/docs/workroom) explains how editor and agent work fit together. - [Full Auto](/docs/full-auto) explains the current automation direction and its proof limits. - [Review and evidence](/docs/review-and-recovery) explains review and recovery principles. - [Security and privacy](/docs/security-and-privacy) explains authority, data, and service boundaries. - [Troubleshooting](/docs/troubleshooting) gives safe checks for current candidates. ## Legacy documentation Earlier public docs described a separate Electron application. That product content is retired from this documentation set. The current public product documentation is for Omega. --- # Availability and setup Source: https://openagents.com/docs/getting-started ## Current availability Omega is in active development. OpenAgents has produced signed and notarized macOS ARM64 prerelease candidates. The current evidence does not prove a generally available or supported release. The latest public observation record still identifies open brand, safety, and installed-journey gates. Do not use a prerelease tag as proof that these gates passed. ## Evaluate a candidate 1. Open the [Omega releases](https://github.com/OpenAgentsInc/omega/releases) page. 2. Read the notes for the exact candidate. 3. Confirm that the artifact matches your operating system and architecture. 4. Confirm the recorded digest and signing status before you run the artifact. 5. Keep Omega data separate from any Zed installation. 6. Report a result against the exact Omega version and artifact digest. A candidate can be useful for evaluation when a release gate is still open. Do not present that candidate as a stable installer. ## Work from source The [Omega repository](https://github.com/OpenAgentsInc/omega) owns the current source build, test, and package commands. Use the instructions at the exact source commit that you test. This website does not copy changing repository commands into a second authority. Omega is a fork of Zed. The fork keeps native Rust and GPUI editor foundations. Omega must use its own application identity, data roots, credentials, service boundaries, package records, and release process. ## Before you report a result Include these values: - Omega version and source commit - Artifact name and digest, when applicable - Operating system and architecture - The action that you performed - The result that you observed - Any open gate or limitation that affected the result Remove credentials, private repository data, local paths, and private agent content from public evidence. --- # Native workspace Source: https://openagents.com/docs/workroom Omega is an IDE, not a wrapper around one agent or one command-line tool. The native workspace is where project work, agents, code, review, and evidence meet. ## Editor and project truth The Zed foundation owns editor, project, buffer, language, terminal, and worktree state. Omega must not create a second project graph for agent work. An agent works against the project that the user opened. File changes, terminal work, and review stay connected to that project context. ## Agent work Omega is being built to show human and agent work in one inspectable timeline. The target surface includes: - User and agent messages - Plans and current work - Tool and command outcomes - Questions, approvals, and interruptions - File changes and review evidence - Delegated agent activity External agents keep their own configuration and credential custody. Omega must not become a second home for those credentials. ## Review and evidence The workspace should connect a result to the code, tools, tests, and evidence that produced it. A status label or generated summary is not proof by itself. Omega keeps release, payment, deployment, and repository authority outside a UI projection. A view can explain an action without gaining authority to perform it. ## Current status Native workspace work is active in the Omega repository. Prerelease candidates and component proofs show parts of this direction. They do not prove that every workspace capability is complete or supported. --- # Full Auto Source: https://openagents.com/docs/full-auto Full Auto is the Omega direction for sustained agent work in a project. It is designed to continue useful work while the user can inspect, steer, or stop the run. Full Auto is in active development. Current engine and interface evidence does not prove general release availability. ## Intended experience Full Auto should: - Use the project and execution context that the user selected. - Keep agent work visible in the same workspace. - Show plans, tools, changes, blockers, and delegated work. - Preserve one causal record across continuations. - Stop when a safety, authority, or evidence gate requires a decision. - Make interruption and recovery explicit. It must not create a hidden workroom or convert a UI control into deployment, payment, repository, or release authority. ## Agent routing Omega is not limited to one agent provider. The current direction supports a first-party Omega agent and external agents through explicit routing boundaries. Each agent keeps its own configuration and credential custody. Omega records the route and result without claiming that one provider owns the project. ## Safety boundary Automation can act only within the authority that the selected environment already grants. A running agent does not gain more authority because Full Auto is on. The product must keep these states visible: - Running work - A pending question or approval - A failed or interrupted continuation - An unavailable provider or engine - A completed result that still needs review ## Proof status OpenAgents has current Full Auto contracts, engine work, and proof tooling. Installed replay, sustained owner use, and independent release evidence still have open gates. Treat Full Auto as a development capability until one exact candidate passes those gates. --- # Review and evidence Source: https://openagents.com/docs/review-and-recovery ## Review in project context Omega should keep review next to the editor and the agent work that produced a change. The review surface must identify the project, files, and current worktree state. Review can show: - Exact file changes - Commands and test outcomes - Agent decisions and delegated work - Evidence references - Known gaps and failed checks A generated summary is not a substitute for the underlying change or result. ## Authority boundary A review view does not gain Git, deployment, payment, or release authority. Each action still uses its owning system and its current authorization check. Stale, secret-shaped, oversized, binary, or unavailable content must fail visibly. The product must not copy private credentials or unrestricted host data into an agent timeline. ## Recovery Omega should reconcile an interrupted run against its durable identity and evidence. It must not silently rerun provider work after a restart. Use this recovery sequence: 1. Confirm the project and worktree. 2. Confirm the selected run or conversation. 3. Read the latest interruption or failure. 4. Check the exact file and test state. 5. Resume only through a visible action. 6. Start new work only when you intend to create a new run identity. If Omega cannot prove the current state, it must show that the state is unknown or needs reconciliation. ## Current status Review and recovery are active product areas. Current prerelease evidence does not prove every review, restart, and recovery path on a supported release. --- # Security and privacy Source: https://openagents.com/docs/security-and-privacy ## Native authority Omega is a native application. The host process owns project and local execution access. UI projections must receive bounded data and typed actions, not unrestricted credentials or host authority. An agent action uses the authority of its selected environment. Omega does not grant repository, deployment, payment, or release authority by displaying that action. ## Application identity and data Omega must use its own application identity, data directories, logs, cache, credentials, protocol handlers, and update state. It must not read or change Zed user data as part of normal Omega operation. Prerelease evaluation should use the candidate-specific Omega identity. Do not point an Omega candidate at a Zed data directory. ## Credentials Provider and account credentials stay with their owning system. External agents keep their own configuration and credential custody. Do not put credentials, private prompts, repository secrets, private transcripts, or unrestricted local paths in logs, screenshots, documentation, or public evidence. ## Inherited services Omega is a fork of Zed, but it is not authorized to present Zed services as Omega services. A release candidate must disable inherited production endpoints by default or identify an approved compatibility use. An unavailable Omega service must have an honest state. It must not silently call an inherited account, telemetry, crash, update, or hosted-model endpoint. ## Release boundary Signed and notarized bytes are necessary release evidence. They do not prove that privacy, network, data isolation, accessibility, or installed-journey gates passed. The current public Omega evidence still has open release gates. Do not infer stable support from a prerelease artifact. --- # Troubleshooting Source: https://openagents.com/docs/troubleshooting ## A candidate does not start Confirm the exact Omega version, artifact digest, operating system, and architecture. Read the candidate release notes before you change system policy. Do not disable signing, quarantine, or security controls to make an unknown artifact run. Report a package or notarization failure against the exact artifact. ## The product shows Zed as the product Record the Omega version, visible surface, and exact text or image. A retained internal compatibility identifier can be valid. User-facing Zed product, service, package, or data text can still be a release blocker. Do not run an uninstall or migration command when its target is unclear. ## An agent or Full Auto is unavailable Confirm the selected project, agent, provider, and execution environment. Read the visible failure before you retry. Do not replace a failed provider with another provider unless you intend to change the route. Do not treat a source test or fixture as proof that an installed candidate completed the same journey. ## Review or evidence is incomplete Confirm the worktree and candidate identity. Then inspect the exact changes, commands, tests, and open gates. Do not convert an unknown, waived, or unperformed check into a pass. ## Report a reproducible issue Open an issue in the [Omega repository](https://github.com/OpenAgentsInc/omega/issues). Include: - Omega version and source commit - Artifact name and digest, when applicable - Operating system and architecture - Reproduction steps - The result that you expected - The result that you observed Remove credentials, private repository data, local paths, and private agent content before you attach evidence. --- # Agent-readable docs Source: https://openagents.com/docs/agent-readable The docs build emits static agent-readable artifacts. It does not run an AI endpoint or an MCP server. ## Available artifacts - [`llms.txt`](https://openagents.com/docs/llms.txt) indexes the public documentation tree. - [`llms-full.txt`](https://openagents.com/docs/llms-full.txt) provides the complete public documentation corpus. - [`agent-readability.json`](https://openagents.com/docs/agent-readability.json) describes enabled static formats and content-use signals. - Add `.md` to a documentation route to request its raw Markdown mirror. These artifacts contain only the curated public content in this docs package. They do not index internal audits, runbooks, private traces, operational topology, or the complete repository `docs/` tree. ## Authority boundary Documentation helps a tool understand Omega. It does not grant authentication, repository, deployment, payment, moderation, or release authority. Search uses a local static index that the site builds. Ask AI and hosted MCP are not part of this docs surface. ## Availability language Omega is in active development. A signed prerelease candidate, source test, screenshot, or producer record does not prove general availability. Agents must read the version, candidate digest, result, open gates, and reviewer status together. The [Omega releases](https://github.com/OpenAgentsInc/omega/releases) page contains candidate artifacts and notes. The OpenAgents repository contains supporting plans and evidence records. ## Product claim authority Product promises are machine-facing evidence. Agents that need current claim state should read the structured projections: - [`/api/public/product-promises`](https://openagents.com/api/public/product-promises) is the current registry projection. - [`/api/public/product-promises/transitions`](https://openagents.com/api/public/product-promises/transitions) is the transition receipt stream. - [`/api/public/product-promises/audit`](https://openagents.com/api/public/product-promises/audit) is the audit projection. Read the promise identifier, state, evidence, blockers, and caveat together. Documentation does not independently make a capability available. `/docs/product-promises` remains a compatibility redirect to this agent-facing boundary.