Workspai gives developers and AI agents a shared, evidence-backed view of a software workspace, then helps them investigate, change, and verify it from inside VS Code.
Ask → inspect → change → test → verify
- Install Workspai.
- Open an existing repository or start a new workspace in VS Code.
- Open the Workspai sidebar, then adopt or create your project.
- Ask a question, prepare a plan, run an Agent task, or define a Goal.
The extension includes a verified Workspai CLI runtime. You do not need a separate global CLI installation for extension-owned operations.
| Mode | Best for |
|---|---|
| Ask | Understanding code, architecture, diagnostics, and workspace state. |
| Plan | Investigating a task and preparing a grounded, read-only plan. |
| Agent | Editing code, running scoped checks, and verifying the result. |
| Goal | Working toward a defined outcome until its required checks pass. |
Ask and Plan stay read-only. Agent and Goal can make changes, but every write is scoped, reviewable, and verified.
Workspai combines source code with projects, dependencies, contracts, diagnostics, documentation, and operational evidence. This gives the assistant a focused map of the system before it decides where to reason or what to change.
Describe the workspace or project you want. Workspai prepares a clear plan for review before it creates files or runs setup commands.
Use one conversational loop to inspect evidence, edit source, run project commands, review changed files, and verify the outcome. When an operation needs permission, Studio shows a compact approval card in the conversation. You can run it once, allow the exact action for the session or project when supported, or decline it.
An ordinary Agent task does not need an incident card. It reads before changing, then verifies the result against fresh evidence.
Explore how projects, modules, files, APIs, and dependencies connect. Select an entity to inspect its relationships and supporting evidence.
See what Workspai and its agents are doing: active commands, completed stages, failures, verification, model usage, and measurable context efficiency. Operations Floor correlates that activity with the bounded canonical Graph, tool attempts, PCC assurance, and evidence provenance in one presentation-ready view.
Open the generated Model, Graph, agent context, Skills, reports, and other workspace evidence without searching through internal directories.
- Read-only investigation can run without unnecessary prompts.
- Source changes are scoped, recorded, and reversible where supported.
- Sensitive or invasive actions require an explicit, exact approval.
- Agent completion requires verification, not just a plausible answer.
- Agent changes can carry their intent, observed effects, and verification as one inspectable record.
Use VS Code models, a local model, or your preferred provider:
VS Code Models · OpenAI · Claude · Gemini · Kimi · DeepSeek ·
OpenRouter · Groq · Mistral · xAI · Ollama · Custom
Provider credentials are stored in VS Code Secret Storage.
- Getting started
- CLI Repair Engine integration
- Release notes
- Workspai CLI
- Contributing
- Issues and feature requests
MIT © Chistiq