Codex Sedna

Codex Sedna is the SednaLabs downstream fork of Codex CLI. It stays close to the upstream OpenAI Codex experience while shipping Sedna-owned release artifacts, downstream validation policy, and fork-specific runtime behavior.

If you are looking for the upstream OpenAI distribution, IDE integrations, or Codex Web, use the official Codex documentation and chatgpt.com/codex. The upstream npm install -g @openai/codex and brew install --cask codex paths install the OpenAI distribution, not Sedna-owned release artifacts.

Install

Use the latest sednalabs/codex GitHub Release for supported Codex Sedna binaries.

Current public Sedna releases publish one supported CLI archive:

• Linux x86_64: codex-sedna-<version>-x86_64-unknown-linux-gnu.tar.gz

The archive contains codex plus codex-responses-api-proxy. macOS, Windows, Linux arm64, and other historical upstream targets remain in the source tree for future re-enablement, but they are not currently supported Sedna release targets.

To build from source, follow Installing and building.

Run codex and select Sign in with ChatGPT to use Codex through your ChatGPT plan. API-key usage is also supported with the upstream Codex auth setup described in the official Codex docs.

What Sedna Adds

Sedna keeps the fork intentionally narrow in hot upstream code and moves downstream behavior behind explicit seams where possible. The main additions are:

Area	What changes in this fork
Release ownership	Sedna publishes its own GitHub releases, version naming, and Linux x86_64 release policy.
Validation	Heavy validation and buildability checks are GitHub-first, with lane docs for targeted, frontier, and release workflows.
Native computer use	The open-source Sedna adapter layer promotes bare Android, browser, and desktop tool names to Codex-owned native computer-use tools with transcript, app-server, TUI, rollout, and trace support.
Browser use	The repo includes a built-in Playwright browser provider for local Chrome/Chromium; command providers can supply signed-in Chrome, in-app-browser, remote, or hosted browser runtimes.
Model-visible screenshots	Successful visual browser, Android, and desktop observations must return native image content to the model. Artifact paths are diagnostics, not the primary visual channel.
Agent orchestration	Downstream carries blocking wait patterns, richer sub-agent inventory, usage/accounting projection, and continuity improvements for long-running local workflows.
Runtime accounting	The fork maintains first-party local usage/accounting surfaces so live CLI, TUI, and app-server views can explain active-thread and combined-session usage.
MCP and config safety	Downstream preserves safety controls around MCP configuration, OAuth fallback behavior, headless device login with dynamic client registration, approval memory, and related runtime guardrails.

For the full inventory, read Downstream / fork notes and the Downstream regression matrix.

Headless MCP Device Login

Codex Sedna supports the OAuth 2.0 Device Authorization Grant (RFC 8628) for configured HTTP MCP servers that advertise a device authorization endpoint in their OAuth metadata. This is useful in SSH sessions, remote shells, CI runners, and other headless environments where a localhost browser callback is inconvenient.

codex mcp login <server-name> --device-auth

Codex performs dynamic client registration when the server supports it. If the server does not support dynamic registration, configure the server entry with a client ID using codex mcp add --oauth-client-id <client-id>.

During login, Codex prints the verification URL and user code from the authorization server, then stores the resulting MCP credentials for the configured server. MCP servers built with the mcp-toolkit-rs hosted HTTP auth surface can expose the metadata needed for this flow.

Native Computer Use

Codex Sedna treats native computer use as a Codex-owned contract backed by replaceable runtime providers.

Codex owns the model-facing tools, transcript events, app-server routing, TUI projection, rollout persistence, and native image-output requirement. Providers own browser sessions, Android sessions, desktop permissions, screenshots, UI digests, input execution, and backend-specific setup.

The open-source implementation in this repository covers the Codex-side computer-use adapter layer and the built-in Playwright browser provider. Native macOS desktop providers, Windows in-app-browser shells, signed-in Chrome extension providers, remote browsers, and Android device providers plug in through documented provider seams instead of being hard-coded into Codex core.

The stable native tool surface includes:

• browser_observe and browser_step
• android_observe, android_step, and android_install_build_from_run
• desktop_observe and desktop_step

The browser bridge supports backend hints such as auto, browser, chrome, chromium, and provider-declared backends such as iab. The built-in Playwright provider can run headless or headed Chrome/Chromium and supports bounded browser actions, fresh post-action screenshots, compact page metadata, selector diagnostics, service-account navigation headers, optional audit artifacts, and per-thread browser profile isolation. External command providers can implement signed-in Chrome, in-app-browser, remote-browser, or hosted browser runtimes behind the same request/response contract.

Read Native computer-use adapter tooling for the runtime contract and Native computer-use cleanroom contracts for the sanitized provider requirements used for macOS desktop, Windows/browser shell, Chrome-extension, bundled-plugin, and Android provider work.

Working With The Fork

This repository has two public branch roles:

• origin/main: maintained downstream branch and public PR target.
• origin/upstream-main: fast-forward mirror of upstream/main.

Completed work on a feature, bugfix, docs, or cleanup branch should be committed, pushed, and opened as a pull request targeting origin/main before handoff. Do not leave finished branch work only in a local worktree.

Downstream syncs are merge-based from upstream-main into main. The intended long-term shape is minimal fork carry in high-churn upstream files, with downstream behavior expressed through narrow extension seams or provider boundaries when possible. Some carry is deliberate product behavior rather than temporary merge residue; the carry ledger and divergence registry mark those surfaces so sync work preserves them until upstream offers an equivalent contract.

Release and validation work is intentionally GitHub-first. Public release artifacts are produced by Sedna release workflows; preview and validation runs are buildability and regression evidence, not release artifacts.

Documentation Map

Need	Start here
Fork identity, branch policy, and divergence inventory	Downstream / fork notes
Release naming, release artifacts, and supported targets	Sedna release policy
Native browser, Android, and desktop computer-use contracts	Native computer-use adapter tooling
Cleanroom provider requirements	Native computer-use cleanroom contracts
GitHub-first validation lanes and remote offload	GitHub CI offload
Multi-step validation workflow	Validation workflow
Downstream tool and regression coverage	Tool surface matrix and Regression matrix
Installing or building from source	Installing and building
Local memories behavior	Memories
Contributing guidelines	Contributing

License

This repository is licensed under the Apache-2.0 License.