CodeAI Hub is a Visual Studio Code extension + standalone Project Manager (CEF) that unifies multiple AI providers behind a single, type-safe orchestration layer.
Current Release — v1.2.321 (VSIX package-size hotfix)
This hotfix restores the VSIX package boundary. The extension package excludes
local Swift native-helper build outputs under native/**, and the release
build now fails if extension/native/, Swift .build caches, ModuleCache,
or dSYM debug artefacts leak into the packaged VSIX. CodeAI Hub remains a
lightweight extension installer/distributor while runtime binaries continue to
ship through the external release artefacts.
The previous Development Tree acceptance fixes remain included.
This release completes the Diagram Modules Development Tree materialization
flow. Core now creates the accepted tree both under
.codeai-hub/<workspace>/development_tree/materialized/... and under
doc/TODO/stages/development-tree/..., so later TODO plans and agent artifacts
have the same Product Part / Cluster / Module folder structure available before
Application Skeleton starts.
Clusters and modules both receive workers/ and integration/ artifact
folders. Cluster root folders remain the place for cluster description and
facade-contract drafts, while the nested operation folders hold worker task
evidence and integration handoff artifacts.
Project Manager still remains a projection-only surface: it consumes Core-owned Development Tree snapshots and auto-reveals the first Product Part / Cluster in the left sidebar so Module / Facade Specification, Implementation, Workers, and Integration nodes are visible immediately.
The previous Development Tree materialization behavior remains included.
The previous session context and Kimi usage telemetry behavior remains included.
This hotfix keeps Kimi selected across managed workflow steps. When Description
or an upstream managed stage is started with Kimi, the next start card now
inherits kimiCode instead of falling back to Claude as the first connected
provider.
The previous Kimi progress summary behavior remains included. The managed Kimi profile asks Kimi to avoid full detailed reasoning output as a work log and to compress analysis into short visible ordinary assistant summaries when possible.
The previous managed profile behavior remains included. Kimi starts through a
CodeAI-owned managed agent profile instead of
the provider default agent prompt. The runtime materializes
~/.codeai-hub/providers/kimi/home/codeai-managed-agent/ and passes
--agent-file, an empty --mcp-config-file, and an empty --skills-dir before
starting Kimi Wire.
The managed Kimi profile omits AGENTS/project instruction injection and provider skills from the system prompt, keeps CodeAI Core as the workflow prompt authority, and narrows the available tools to file read/search/write/edit operations. It also asks Kimi to send visible ordinary assistant progress updates during long managed turns so hidden reasoning does not leave the user staring at a silent session.
The previous Kimi reasoning display behavior remains included. Kimi think
content respects the Kimi Reasoning in dialog setting, renders as an expanded
thinking bubble instead of the retired collapsed panel, and streams bounded
reasoning chunks before long turns finish when Kimi Wire provides intermediate
thinking content.
The previous Kimi review input unlock behavior remains included. Kimi TurnEnd
normalizes to a Core-compatible turn_completed event with
postTurnTokenUsageUnavailable=true, allowing Core continuity arbitration to
resolve the post-turn check as no-rollover and return the runtime session to
idle after user-review cards.
Installed Kimi Code / Kimi 2.6 sessions also apply the Core-provided workspace
before Wire startup. When Core calls createSession(workspacePath), the Kimi
adapter rebuilds its runtime configuration before the first Wire process
starts, so both --work-dir and the process cwd point at the actual project
workspace instead of the early Core launcher working directory.
Kimi Wire ContentPart text and thinking chunks are normalized into
Core-compatible assistant and thinking messages, buffered per turn, and
flushed before turn_completed, so the dialog history receives the provider
answer and the input panel can leave resume/bootstrap state.
The Kimi module also resolves the user-local kimi CLI from KIMI_CLI_PATH,
~/.local/bin, Homebrew paths, or the inherited PATH, and uses string
JSON-RPC request ids required by Kimi Wire. Kimi sessions also pass the selected
workspace through --work-dir and use protocol-compatible approval literals for
managed workflow artifact turns.
The Kimi provider module remains available across CodeAI Hub. Kimi
uses Wire mode through the standalone packages/Kimi_Module facade, keeps
CodeAI-managed runtime state under ~/.codeai-hub/providers/kimi/home via
KIMI_SHARE_DIR, and references the already authorized Kimi config through
~/.kimi/config.toml.
Kimi is now available in Project Manager settings, Description provider
selection, workflow start/fix cards, provider color mapping, and Session UI
status/model display. The first release exposes kimi-for-coding as the default
model and treats live model switching as display-only until a provider-native
Wire switch contract is verified.
Release packaging now builds kimi-module-<version>.tar.bz2, includes Kimi in
the Core runtime dependency bundle, validates installed Kimi artifacts during
release, and keeps Kimi provider workspace files out of the VSIX.
Project Manager trunk step markers now come from Core-owned workflow state:
gray before a step starts, yellow after Core opens the step session or sends the
first provider prompt, and green only after the stage reaches its terminal
completion boundary. Managed technical stages also pass a terminal clean-Git
checkpoint before publishing green completion; classified residue is committed by
Core, while unclassified dirty files block the next stage until resolved.
Description and Virtual Simulation also promote from yellow to green when their
final artifacts make the next step available, and Diagram Modules flow sidecars
are committed automatically at terminal completion.
Managed commits now stage dot-directory sidecars without Git exclude pathspecs,
so module-map.flow.json is saved by Core instead of blocking the user.
Terminal completion now also treats non-semantic Core metadata as managed
residue, ensures .codeai-hub/state/ is ignored as local runtime state, and
keeps task timer telemetry out of Git history.
Repository lifecycle tooling under scripts/plan-orchestrator/** remains in
place because it powers npm run plan:* and the Husky plan hooks for this code
base. It is separate from the retired generated user-workspace orchestrator.
CodeAI Hub is already usable, but the current recommended installation path is still source-based. If you want to try the product today, clone the repository, build the release artifacts locally, and install the generated VSIX into Visual Studio Code.
- Git
nvm- Node.js per the project
.nvmrc(currently22.17.0) +npm - Visual Studio Code
cmake(required for the standalone CEF launcher / Project Manager build)- the provider CLIs or SDK access you plan to use (
Claude,Codex,Gemini,Kimi) installed and authenticated separately
git clone https://github.com/OleynikAleksandr/CodeAI-Hub.git
cd CodeAI-Hub
nvm install # reads .nvmrc and installs the pinned Node version
nvm use
npm install
npm run setup:hooks
./scripts/build-all.sh
./scripts/build-release.sh --use-current-version- VSIX package in the repository root:
codeai-hub-<version>.vsix - fresh runtime tarballs in:
doc/tmp/releases/~/.codeai-hub/releases/
Open Visual Studio Code and run Extensions: Install from VSIX..., then select the generated codeai-hub-<version>.vsix.
- This is the current early-access path, not a polished one-click installer.
- The first full build can take a while because it prepares provider bundles, UI bundles, core runtime, and the standalone launcher.
- Provider CLIs / SDKs are not bundled inside this repository and must be available separately.
Before starting, read doc/SolidWorks-WorkFlow/Docs_Index.md and follow the SSOT contracts in doc/SolidWorks-WorkFlow/Contracts/ (especially doc/SolidWorks-WorkFlow/Contracts/Workflow_CLI.md) to configure provider CLIs and SDKs.
- Install dependencies
npm install npm run setup:hooks # installs Husky git hooks - Implement changes in
src/andpackages/**(micro-classes + facades; keep files under 500 lines). - Run quality checks before committing:
npm run quality # architecture gate + Ultracite lint npm run check:knip # detect unused files/exports npm run compile # ensure TypeScript builds cleanly
- GitHub Actions now runs a minimal public CI baseline on every push to
mainand on every pull request. - The workflow enforces the same root quality gates used as the local baseline:
npm run check:architecture,npm run lint,npm run check:knip, andnpm run compile. - The root
compilegate now builds@codeai-hub/translation,@codeai-hub/localization, and@codeai-hub/core-supervisorbefore browser/root type-check, so clean GitHub runners do not depend on pre-existing workspacedist/folders. - Local Husky hooks remain the fastest feedback path; CI is the public verification surface, not a replacement for the local release ritual.
./scripts/build-all.sh
./scripts/build-release.sh --use-current-versionmedia/ Bundled webview assets (CSS + JS) shipped with the extension.
media/react-chat.js React bundle generated by the webview build script.
src/extension.ts VS Code extension entry point.
src/extension-module/ Extension host micro-classes (settings, bootstrap glue).
src/core/webview-module/ HTML scaffold that injects the webview assets.
src/client/project-manager/ Project Manager CEF UI surface (sidebar, sessions, settings).
src/client/ui/ Shared session UI bundle (used by both webview and PM shells).
src/types/ Shared TypeScript types (provider, session, model registries).
packages/ Workspace packages — provider modules and runtime services.
packages/Claude_Module/ Claude provider runtime (Agent SDK integration, session lifecycle).
packages/Codex_AppServer_Module/ Codex provider runtime (App Server JSON-RPC, app-server process).
packages/Gemini_Module/ Gemini provider runtime (CLI core integration).
packages/core/ Core orchestrator (turn lifecycle, continuity, remote bridge).
packages/core-supervisor/ Runtime supervisor (Core process management).
packages/cef-launcher/ Standalone CEF Launcher (native macOS/Windows/Linux client for PM).
packages/ui/ Project Manager UI styles + bundled assets.
packages/translation/ Shared runtime translation engine.
packages/localization/ Bundled English source dictionaries + glossary + lookup primitives.
packages/unified-session/ Shared session contract used across packages.
scripts/ Quality, build, and release automation (build-all.sh, build-release.sh, etc.).
doc/ Architecture SSOT, planning docs, sessions log, knowledge base.
doc/SolidWorks-WorkFlow/ Canonical SSOT tree (System / Clusters / Modules / Contracts / DesignSystem / Plans).
doc/tmp/releases/ Locally staged release tarballs after build-all.sh.
This repository is currently distributed as UNLICENSED. Source is visible for audit and development collaboration, but redistribution requires explicit permission from the repository owner.