OpenVibez

OpenVibez is a local-first desktop AI coding assistant built with Electron, React, and TypeScript.

It connects directly to the providers you already use, stores secrets in your OS keychain, keeps app data in local SQLite, and runs provider and shell orchestration in the Electron main process instead of a hosted relay.

Status: Early alpha.

Listen bub: this is sketchy, experimental, vibey software. It can run local shell commands, supports elevated/root-style execution modes, and absolutely should not be treated like a polite little sandboxed toy. If you point it at something important, do that with your eyes open.

What It Does Today

• Direct provider connections with no OpenVibez cloud relay
• Local persistence for sessions, messages, settings, usage, and run recovery state
• OS keychain secret storage via keytar
• Streaming chat with live status updates and trace/timeline output
• Workspace attachment with trust levels and scoped vs root execution modes
• Command policy enforcement for high-risk commands, read-only workspaces, and out-of-scope cwd access
• Per-provider model sync and selection
• Markdown rendering with syntax highlighting in chat messages
• 30-day token and cost summary in Settings

In plain English: it talks straight to model providers, keeps most of its mess on your machine, and is trying very hard to be useful before it is respectable.

Provider Support

Provider	Status	Notes
OpenAI API	Shipped	Native OpenAI path, model sync, custom base URL / OpenAI-compatible endpoints, optional background-mode pilot
ChatGPT subscription	Shipped	Codex device login flow, approval policy controls, optional output schema, Codex SDK pilot fallback path
Anthropic	Shipped	Native provider connection plus autonomous local CLI execution flow
Gemini	Shipped	Native Gemini provider path and model sync
OpenRouter	Shipped	Native OpenRouter provider path, provider headers, pricing sync, cost attribution, autonomous local CLI execution with direct-response fallback
Grok (xAI)	Shipped	Native Grok provider path with direct xAI API support and autonomous local CLI execution
Ollama	Shipped	Local models, diagnostics, runtime controls, native tool-calling path with protocol fallback

Feature Breakdown

Core app

• Electron desktop app with React renderer and typed preload IPC bridge
• Main-process ownership of secrets, DB access, provider calls, background jobs, and shell execution
• SQLite storage for providers, model profiles, workspaces, sessions, messages, usage events, settings, assistant runs, and background jobs
• Automatic session title generation from conversation context

Chat and agent flow

• Streaming assistant output with cancellable in-flight runs
• Trace rendering for plan, thought, action, command, and command-result events
• Per-session timeline persistence in app settings
• Restart reconciliation for interrupted runs, plus OpenAI background polling for long-running responses

Workspace and execution controls

• Attach local projects as workspaces
• Trust levels: trusted, read_only, untrusted
• Access modes: scoped and root
• Policy gates:
  • scoped commands must stay inside the selected workspace
  • read_only workspaces block mutating commands
  • untrusted workspaces block shell execution
  • when a workspace is selected, root mode requires it to be trusted
  • known high-risk commands are blocked

Provider-specific extras

• OpenAI-compatible endpoint profiles with saved defaults
• OpenRouter app origin/title controls and synced pricing map
• Ollama diagnostics plus temperature, output-token, and num_ctx controls
• Codex approval-policy and output-schema controls

Project Layout

openvibez/
├── apps/desktop/     # Electron app (main, preload, renderer)
├── db/               # Drizzle schema + migrations
├── docs/             # specs and architecture notes
└── packages/shared/  # shared contracts/types

Getting Started

git clone <repo-url> OpenVibez
cd OpenVibez
npm install
npm run dev

Other useful commands:

npm run build
npm run typecheck
npm run db:generate
npm run db:studio

Provider setup

1. Open Settings.
2. Add a provider.
3. Save a key or complete ChatGPT device login.
4. Run Save + Test or Check to sync models.
5. Start a session and pick a model/access mode.

If everything boots first try, act normal. Nobody likes a showoff.

Environment variables

Variable	Description
OPENVIBEZ_DEVTOOLS	Set to 1 to auto-open DevTools on launch
OPENVIBEZ_CODEX_BIN	Absolute path to the Codex CLI binary for subscription mode

Roadmap

Highest priority

• Code block actions: copy, save/apply-to-file, and tighter code review ergonomics
• Conversation search, filtering, and better session management
• System prompt controls per session or workspace
• Better session branching/forking workflows

Provider and runtime work left

• Provider health/status checks for stale keys, broken endpoints, and degraded local runtimes
• Broader provider-native tool use instead of compatibility/prompt-protocol fallbacks where possible
• Stronger long-run recovery and resume behavior outside the current OpenAI background pilot
• More explicit per-provider presets and model defaults

Shipping and hardening

• Packaged release workflow and auto-update support
• Opt-in crash reporting / diagnostics
• Headless or CLI companion mode
• Extension/plugin surface for custom providers or tools
• Automated test coverage for provider flows, restart recovery, and command-policy enforcement

Tech Stack

Layer	Stack
Desktop shell	Electron 39
UI	React 19, Vite 7, TypeScript 5.9
Styling	Tailwind CSS, Radix UI
State	Zustand
Persistence	SQLite + better-sqlite3, Drizzle ORM
Secrets	keytar / OS keychain
Providers	OpenAI SDK, Codex CLI / SDK pilot, direct provider integrations

License

MIT

Ya I also vibed the README.