What is Shrevix
Shrevix is a coding + knowledge agent in a desktop app. You give it an instruction; it picks the right capability — reading and editing code, running commands, searching your private knowledge or the web, looking at an image — and shows every step as it goes.
What makes it different
Section titled “What makes it different”- Subscription, built-in, and compatible providers. Codex Local can use an existing signed-in Codex subscription, while DeepSeek, Gemini, OpenRouter, OpenCode Zen, Kimi, GLM, NVIDIA, Ollama, and LM Studio are registered in the app. Claude is available through OpenRouter, and custom OpenAI-compatible endpoints can be added.
- Approval-gated by default. File edits show a diff first. Shell commands and MCP calls wait unless the selected mode or an explicit MCP allowlist says otherwise. The gate is enforced in code, not prompts.
- Local-first. State lives on your disk and knowledge embeddings are computed locally on CPU. A remote model still receives the prompt and selected context needed to answer; a local model keeps inference on-device.
- An org, not a bot. Agents form teams with leads that inspect the roster, route to the right specialist, and run independent assignments in parallel. Each child keeps a separate, openable conversation.
- Separate workspaces. Each task can use its own git worktree and branch. This separates task changes; shell execution is approval-gated host execution, not an OS security sandbox.
A tour of the window
Section titled “A tour of the window”Everything lives in one window, organized down the sidebar:
| Page | What it’s for |
|---|---|
| Dashboard | Greeting, new-task composer, Needs you queue, recent sessions |
| Sessions | Searchable, resumable conversation history |
| Repositories | Onboard codebases from GitHub or a local path |
| Workspaces | Isolated worktrees, one per task |
| Agents | The org chart: teams, leads, agent profiles |
| Workflows | Durable multi-step definitions, triggers, runs, and human gates |
| Skills | Reusable Markdown playbooks the agent can load on demand |
| Knowledge | Your ingested documents, stats, and semantic search |
| Documents | Plans, PRDs, and reports your agents produce — versioned, commentable |
| Hooks | Observe, block, or rewrite events at explicit run stages |
| Connections | Search and authorize external services and MCP integrations |
Alongside every chat you can open tabs: editable files, Markdown preview, quick file and workspace search, named terminals, code viewers, and diff views — all scoped to the conversation’s workspace.
Where to go next
Section titled “Where to go next”Install the app, set up a model key, then run your first conversation.