A native desktop workspace for durable Codex, Claude Code, and local-model sessions. Isolate work in git worktrees, inspect every diff, run tests and browser checks, and recover the context normal terminals lose.
UX · billing redesign
claude-code
Backend · entitlement cache
codex/openai
Database · migration 0042
codex/openai
Review · PR #142
claude-code
The problem
Terminal tabs close. CLI state vanishes. Resume IDs disappear. Recreating context costs the whole afternoon.
Chat lives here. Diff lives there. Tests, browser, notes, terminal — all in different windows you tab between.
Agents produce diffs faster than humans can read. Without prompt, diff, and tests in one place, review breaks down.
Multiple developers and agents touch the same repo. Worktrees and branches drift without anyone noticing.
The flow
Every step is part of the product surface — not a separate tool you have to context-switch into.
Track local projects, repository metadata, commands, runtime defaults, and worktree policy.
Choose Codex/OpenAI, Codex/Ollama, Codex/LM Studio, custom OpenAI-compatible, or Claude Code.
Default to a safe worktree session. Explicit opt-in for main repo. Base branch and commit always visible.
Chat, terminal, diff, editor, test runner, browser, Playwright, MCP, hooks, and audit — all session-scoped.
Prompt markers, execution diffs, git status, comments, test output, and validation events stay connected.
Create local checkpoint commits, prepare PRs through gh, or route findings into the next fix session.
Reconnect to the daemon and replay persisted session history when the UI or runtime fails.
The product
Six surfaces, one cockpit. Each piece earns its real estate by sitting next to the agent turn that needs it.
All non-terminated sessions stay visible per project, with status, follow-up input, mode changes, stop controls, popouts, and replay.
Start AI work in a scoped lane, then inspect exactly what changed before it becomes a PR.
Validate where the agent worked. Test commands, browser preview, and Playwright runs in the same session surface.
Launch installed agent CLIs locally. Secrets stay in env vars, credential managers, SSH agents, or keychain references.
See who is viewing a project, invite collaborators through GitHub, and open read-only live session popouts.
When the app reconnects, replay persisted history, surface crashed sessions, and keep audit evidence attached.
also included
Terminal · editor · logs · dashboard · setup · todo · PR status · rules · MCP · hooks · audit · roadmap · dependencies · lifecycle · database inspection.
Local-first by design
Vibe Toolbox runs agent CLIs locally, stores project and session state in a local SQLite database, and only talks to the cloud for what the cloud should actually do.
On your machine
Vibe Toolbox Cloud
Team visibility
GitHub-backed identity, invitations, project presence, shared active-session listings, and read-only live popouts. Hosted relay coming with paid Team.
Watch a teammate's session transcript without taking control. Prompt input, stop, and mode changes stay hidden.
Invite collaborators through the same GitHub flow that grants repo access. No manual peer-token exchange.
Coordinated sessions across the internet. Available with paid Team once hosted relay and seats are production-ready.
Team plan launch — `$49/seat/month` (planned)
Early-access design partners: $39/seat/month.
Pricing
Stripe-backed subscription with a 14-day trial. Up to 3 activated devices and 7-day offline grace.
Prove the workflow on your own machine before paying.
For solo vibe coders who want a real workspace around AI coding sessions.
For small teams coordinating AI sessions across the same repositories.
FAQ
No. It is a native AI development workspace. File, terminal, git, diff, test, browser, and review surfaces are included, but it is designed to coordinate AI coding sessions around existing projects — not replace every editor workflow.
Codex and Claude Code. Codex runtime options include OpenAI/default, Ollama, LM Studio, and custom OpenAI-compatible endpoints, with approval policy, sandbox mode, web search, and config overrides. External runtimes require your installed and authenticated CLIs.
By default, no. Source code, prompts, diffs, terminal output, and model responses stay on your machine. The cloud is used only for account identity, subscription entitlements, release access, planned hosted team collaboration, and explicit support diagnostics.
The daemon owns long-running sessions, supervision, SQLite persistence, recovery, git/worktree operations, panels, SSH/tmux orchestration, and structured agent output parsing. That lets the UI reconnect and replay useful history instead of losing everything when a window closes.
Local projects, durable AI sessions, Codex/Claude/local runtime setup, worktrees, git review, native panels, signed updates, diagnostics export, and personal workflow features. Solo is $29.99/month.
Team is on the way. The codebase already has local presence, GitHub-backed invitations, read-only live popouts, and direct trusted peer transport for LAN/VPN validation. Paid Team launches after the hosted relay, seats, permission checks, and production validation are complete.
Windows x64 first, through a signed Tauri desktop app with local Windows updater publishing. macOS and Linux are not promised yet — release scope expands after Windows is stable.
/ ship
Start a 14-day trial of Vibe Toolbox Solo, or join the Team waitlist for hosted collaboration.