Skip to content

Releases: yyjeqhc/webcodex

WebCodex v0.2.0

Choose a tag to compare

@yyjeqhc yyjeqhc released this 07 Jul 13:05

WebCodex 0.2.0

English

WebCodex lets ChatGPT and other MCP/GPT Action clients safely operate your private codebase through a self-hosted, auditable tool runtime.

It turns online AI coding from blind file edits into a permissioned, validated, reviewable engineering workflow.

Highlights

  • ChatGPT, MCP, and GPT Actions can use one self-hosted coding runtime.
  • MCP and GPT Actions call the same WebCodex ToolRuntime.
  • Projects are registered by agents, with runtime ids shaped like agent:<client_id>:<project_id>.
  • The recommended coding workflow is structured: start, inspect, edit, validate, review, finish.
  • Sessions, handoff summaries, finish verdicts, hygiene checks, and validation summaries provide review evidence.
  • Structured edit tools support scoped line edits and checked patch application.
  • Runtime health, project discovery, Git status/diff, Cargo validation helpers, bounded shell/job execution, and artifact workflows are available through the runtime.
  • Legacy server-side projects.toml onboarding, legacy /api/codex/* routes, and run_codex are removed from the model-facing runtime.

Who Should Try This

Try WebCodex 0.2.0 if you want an online AI coding client to work on private code while repository execution stays on a machine you control.

Good fit:

  • Solo developers and small teams comfortable self-hosting a server and agent.
  • Teams evaluating ChatGPT MCP or Custom GPT Actions for private repositories.
  • Operators who want auditable coding tasks with explicit validation and review evidence.

Not yet a fit:

  • Teams looking for hosted SaaS.
  • Teams requiring first-class IDE/LSP semantics in the first setup.
  • Operators who cannot manage token, HTTPS, agent, and shell boundaries.

What Changed

  • Added a productized online coding loop around start_coding_task, structured inspection, structured edits, validation helpers, review/hygiene tools, and finish_coding_task.
  • Improved agent-registered project discovery and runtime project id guidance.
  • Aligned MCP and GPT Actions around the same ToolRuntime and safety model.
  • Added or refined session/handoff/finish summaries for reviewable closeout.
  • Added structured line edit and patch workflows for safer source changes.
  • Added runtime health and compact discovery patterns for model-facing clients.
  • Added OAuth/PAT/shared-key documentation for deployment paths while keeping README and Quick Start focused on first setup.
  • Expanded Chinese onboarding docs for README, Quick Start, Concepts, MCP, and GPT Actions.

Breaking Changes

  • Server-side projects.toml project onboarding is removed.
  • Projects must be registered by agents.
  • PROJECTS_CONFIG is not the runtime project source.
  • Legacy /api/codex/* routes are removed.
  • run_codex is removed from the model-facing runtime surface.
  • Operators who need Codex-specific workflows should run them outside WebCodex.

Security Model

  • The model can only call exposed tools.
  • Project access is agent-registered.
  • The server does not scan the filesystem.
  • Shell/job tools are bounded but powerful and should be treated as escape hatches.
  • Tokens, shared keys, env files, Authorization headers, and complete agent configs must not be exposed in prompts, logs, examples, or committed files.
  • Session and finish outputs provide bounded review evidence, not a replacement for normal code review or infrastructure logging.

See ../SECURITY.md and CONCEPTS.md.

Known Limitations

  • WebCodex is self-hosted infrastructure, not hosted SaaS.
  • First setup is still technical and assumes command-line comfort.
  • Semantic code intelligence, LSP diagnostics, references, and symbol outline are not first-class in 0.2.0.
  • The UI/dashboard is minimal and not the primary workflow.
  • Shell/job tools require trust, scoped configuration, and operator discipline.
  • Full production hardening still depends on your HTTPS, token, OS-user, reverse-proxy, and agent deployment choices.

Upgrade Notes

  • Binary users should replace webcodex, webcodex-cli, and webcodex-agent with matching 0.2.0 artifacts when the release artifacts are published.
  • If using an npm wrapper, confirm which binary version its manifest installs before assuming it is 0.2.0.
  • Re-register projects through agents instead of relying on removed server-side project configuration.
  • Refresh GPT Actions schemas from /openapi.json after upgrading.
  • Reconnect or restart agents after changing project registrations, tokens, or shell profiles.

Validation

Before publishing or deploying 0.2.0 artifacts, run the current procedure in RELEASE_CHECKLIST.md. At minimum, verify formatting, cargo check, markdown links, OpenAPI/MCP smoke discovery, agent project discovery, a read-only coding task, and a small reversible edit task.

This release note is not a substitute for fresh validation output from the final release commit.

Next

The 0.2.x line focuses on making the online coding loop easier to try, validate, review, and roll back. See ROADMAP.md for the short roadmap.

WebCodex v0.1.0

Choose a tag to compare

@yyjeqhc yyjeqhc released this 27 Jun 14:52

WebCodex v0.1.0

First binary release for WebCodex.

Supported platforms

  • linux-x64
  • linux-arm64
  • darwin-arm64

Included binaries

Each archive contains:

  • webcodex
  • webcodex-agent
  • webcodex-cli

Notes

  • GPT Actions should use a user API token, not the server bootstrap token.
  • Server/admin setup uses webcodex-cli server ... and webcodex-cli pairing create.
  • Client setup uses webcodex-cli client enroll, webcodex-cli agent install-service, and webcodex-cli doctor --strict.
  • Windows and Intel macOS artifacts are not included in this release.