Home Vinci Code

Vinci Code

By George Pu
• 7 articles

Using slash commands in Vinci Code

Vinci Code supports slash commands that let you control your session, export work, and access settings without leaving the chat. Type any command below to use it. Session and navigation commands - /new — Start a new session - /resume — Resume a previous session - /name — Set or change your session's display name - /session — Show session info and statistics - /fork — Create a new fork from a previous message - /clone — Duplicate the current session at the current position - /tree — Navigate the session tree (switch between branches) Model and configuration - /model — Switch the Vinci class for this session only; it never changes your account setting - /scoped-models — Enable or disable specific models for Ctrl+P cycling - /settings — Open the settings menu Export, import, and sharing - /export — Export your session as HTML (default) or as .jsonl or .html by specifying a path - /import — Import and resume a session from a saved JSONL file - /share — Share your current session as a secret GitHub gist - /copy — Copy the last agent response to your clipboard Context and workflow - /compact — Manually compact the session context - /hotkeys — Show all available keyboard shortcuts - /changelog — Show recent changelog entries - /reload — Reload keybindings, extensions, skills, prompts, and themes Authentication and tools - /login — Configure provider authentication (API keys, credentials) - /logout — Remove saved provider authentication - /trust — Save your project trust decision for future sessions Quit - /quit — Exit Vinci Code How to use Type a slash (/) at the start of a message, then the command name and any options. For example: /export ~/my-work.html Most commands will open a UI or prompt for more information if needed.

Last updated on Jul 28, 2026

How do I use Vinci Code's multi-agent Crew?

Vinci Code can spawn multiple agents (called a Crew) to work together on complex tasks. Each agent is isolated, independent, and can review and integrate the work of others. Spawn a Crew Use the /agents command (or /helpers as an alias) to tell Vinci to create and spawn child agents: /agents "Create three agents: one to write the backend API, one to build the frontend, and one to write tests" Each agent: - Inherits the session's active Vinci class and works from an isolated snapshot of your project worktree - Runs independently with its own reasoning and tool use - Can navigate the codebase, run tests, and make changes to its own copy How agents work together - Independent work — each agent explores and implements without waiting for the others. - Patch review — verified ordinary patches (changes that pass Vinci's review) auto-integrate into the main worktree. - Status visibility — stale, risky, conflicting, or unverified patches show their status so you can see what's pending. - No capacity limits — Crew capacity is operational and never plan-tier gated — spawn as many helpers as the task needs. When to use Crew Crew shines for: - Parallel work streams — front-end, back-end, and testing all happening at once. - Specialized agents — assigning each agent a focused role ("you write documentation", "you optimize performance"). - Large refactors — breaking a big change into independent slices. Monitor and steer Type /task-info to see the current task ID and status. Each agent's work is tracked; you can see progress, review patches, and undo or request changes. For more depth, see the Vinci Code developer guide.

Last updated on Jul 28, 2026

How do I connect Vinci Code without an API key?

Vinci Code includes a keyless login option that pairs your CLI to your Vinci account in the browser — no API key to copy or paste. Install Vinci Code Install the latest release with one command: curl -fsSL https://vinci.getsimpledirect.com/install | sh Then run vinci to start the interactive coding agent. This works on macOS and Linux and needs Node.js v22.19 or later. On Linux, also install bubblewrap (sudo apt install bubblewrap) — Vinci Code runs shell commands inside an OS sandbox and will not run them without it. On Windows, run Vinci Code inside WSL2; Command Prompt and PowerShell are not supported yet. See What do I need to run Vinci Code? for the full list. Pair with your account On first run or when you need to log in: 1. Run vinci and type /login vinci — a browser window opens to platform.getsimpledirect.com/device. 2. Authorize the device with one click. 3. The CLI receives its own revocable key (stored securely in ~/.pi/agent/). 4. You're ready to code — no key to manage or paste. What this means - No secrets in the terminal — the device pairing uses an RFC 8628 flow (the same pattern as GitHub's device login or Google's phone sign-in). - Revocable access — you can revoke the CLI's key anytime on the Vinci Platform under API keys. - Shared billing — the CLI uses your account credits, charged against the same USD balance as Vinci Chat and the mobile app. Using an API key (fallback) If you prefer, you can still use an API key. Generate one on the Vinci Platform, then pass it as an environment variable: export VINCI_API_KEY=your_key_here vinci But device pairing is the recommended way — it's simpler and more secure.

Last updated on Jul 30, 2026

How do I resume an interrupted coding task in Vinci Code?

When Vinci Code is interrupted — your machine sleeps, the network drops, or you close the terminal — you can resume the task safely without losing progress or accidentally replaying commands. Check your task ID Every Vinci Code session has a unique task ID. To find yours, run: /task-info You'll see the task ID and whether it's in progress, done, or waiting for input. Resume a task To pick up where you left off: vinci resume <task-id> Vinci Code will: - Recover the state — knows exactly what files changed and what commands ran - Verify file contents — checks current file state against the last checkpoint - Skip already-done work — refuses to blindly replay an interrupted shell command - Continue the plan — shows your progress and picks up at the right step How recovery works Vinci Code records mutation checkpoints — snapshots of file state — around every write, edit, and bash execution. If you resume: 1. Changed files are verified — Vinci proves what landed on disk from the last checkpoint. 2. Interrupted commands are NOT replayed — if a bash command started but didn't finish, Vinci asks instead of guessing. 3. Progress is preserved — your plan, step counter, and findings all resume where they stopped. Task completion states Every task ends in one of four states: - DONE — code changed and verification passed. Only achievable with explicit confirmation. - DONE-UNVERIFIED — code changed but verification was skipped or incomplete. Flag this if something feels off. - WAITING — Vinci needs a decision from you (e.g., "which approach should I take?"). - BLOCKED — a concrete blocker stopped progress (e.g., missing dependency, authentication required). Undo and recover If Vinci made changes you want to undo: - Type /undo to revert the last turn's file changes (no git needed). - Use /task-info to see recovery status and decide whether to resume or start fresh. For full details, see the Vinci Code guide.

Last updated on Jul 24, 2026

Bring Vinci into your code workflow

When the work is already in a repository, meet Vinci there. Vinci Code is a beta terminal coding agent that can read your project, edit files, run tools, and carry a multi-step task forward. Install and connect Open a terminal and install Vinci Code: curl -fsSL https://vinci.getsimpledirect.com/install | sh You will need macOS or Linux and Node.js v22.19 or later. On Linux, install bubblewrap first (sudo apt install bubblewrap), and on Windows work inside WSL2 rather than Command Prompt or PowerShell — see What do I need to run Vinci Code? for the details. Run vinci to start the agent. At the prompt, enter /login, then connect through browser device pairing with the same Vinci account you use in the app and on the Platform. You will not need to create or paste an API key. Once paired, Vinci Code uses the managed Vinci gateway on whichever model class your account is set to — there is no provider credential to configure, so you can move directly into the project. For code and file tasks, Vinci Forte works best; Vinci Mezzo is the fast, light option for everyday chat, not the recommended class for this workflow. To change the class for every app at once, use Settings → Personalization; to change it for one session only, run /model, which never writes back to your account. The installed CLI uses a static class list, so /model can lag newly available classes. Make the first task easy to review Begin with orientation: “Read this project and explain how the main pieces fit together.” Then choose a bounded change, such as, “Add validation to this form and run the relevant tests.” You’ll see the agent work through the task with a live task list while it reads, edits, and runs tools. Keep your first request narrow enough that you can inspect the resulting diff and test output. If the direction needs adjusting, steer it with another message instead of starting over. Vinci Code maintains checkpoints you can resume from. If you pause or interrupt, you can resume the task from the last checkpoint rather than starting over. That is the first-project rhythm: connect once, give Vinci the context already present in the repository, start with a reviewable change, and use the terminal conversation to guide the work from understanding to verification.

Last updated on Aug 03, 2026

Use Vinci to drive tools in your coding assistant

Use Vinci as the reasoning model inside a coding assistant that already connects to MCP servers. This recipe is for power users who want their current host and tools to stay in place while Vinci decides which tool to call and when. Vinci works with any MCP host that accepts a configurable OpenAI-compatible model. The host remains responsible for connecting to servers and executing tools; Vinci does not run the servers or act as the host. 1. Check your assistant’s provider settings. Continue only if it lets you set an OpenAI-compatible endpoint and model. A host locked to its own model cannot use Vinci for tool selection. 2. In the Vinci Platform, issue a vinci_live_… API key and copy it when shown. Give it the scopes required by your host; inference covers chat-completion calls, while models covers model discovery. 3. Set the base URL to https://vinci.getsimpledirect.com/api/v1. 4. Paste the key into the host’s API-key field and set the model to a class id — mezzo, forte, or fortissimo. For code and file tasks, use forte; Mezzo is fast and light for everyday chat, while Fortissimo is the deepest class for the hardest work. Hosts that require the field to be set should use a class id rather than auto. 5. Connect one or two focused MCP servers in the host. Their tool definitions are passed to Vinci as standard function schemas, so no Vinci-specific adapter is needed. 6. Start with a narrowly scoped request that makes the expected tool obvious. Confirm that the host receives a tool call, executes it, and returns the result. 7. Add more servers or multi-step tasks only after the focused workflow works reliably. On tool-enabled requests, your MCP servers provide the grounding. Plain-chat features such as automatic web search, memory, and attachments do not apply on this path.

Last updated on Aug 03, 2026

What do I need to run Vinci Code? (macOS, Linux, Windows)

Vinci Code is a terminal coding agent, so what it needs is a terminal, Node.js, and an operating system whose sandbox it can use. Here is the whole list. macOS - Apple Silicon or Intel. - Node.js v22.19 or later. Nothing else to install. macOS has the sandbox Vinci Code needs built in. curl -fsSL https://vinci.getsimpledirect.com/install | sh Linux - Node.js v22.19 or later. - bubblewrap. Install bubblewrap before your first session: sudo apt install bubblewrap # Debian, Ubuntu sudo dnf install bubblewrap # Fedora Then install Vinci Code with the same one-line command as macOS. Why bubblewrap is required Vinci Code runs every shell command inside an OS-level sandbox that confines writes to your project folder, so a command that goes wrong cannot reach the rest of your disk. On macOS that sandbox is part of the operating system. On Linux it is bubblewrap, which most distributions do not install by default. If bubblewrap is missing, Vinci Code still starts, still reads your project, and still edits files — but the moment it tries to run a shell command you will see: Vinci disabled bash because no enforceable OS sandbox is available. Install bubblewrap or set VINCI_NO_SANDBOX=1 as an explicit developer bypass. Installing bubblewrap and starting a new session fixes it. We do not recommend the VINCI_NO_SANDBOX=1 bypass outside development: it turns off the layer that keeps commands inside your project. Windows Run Vinci Code inside WSL2 (Windows Subsystem for Linux). Native Windows — Command Prompt or PowerShell — is not supported yet. 1. Install WSL2 by running wsl --install in PowerShell as administrator, then restart. Microsoft's full instructions are here. 2. Open your new Linux terminal (Ubuntu by default) from the Start menu or Windows Terminal. 3. Follow the Linux steps above from inside it — including bubblewrap, which a fresh WSL2 install does not include. Your Windows files are reachable from WSL2 under /mnt/c/, but projects run noticeably faster when they live in the Linux home directory (~/) instead. If you use Git on Windows, clone the repository again inside WSL2 rather than working across /mnt/c/. Checking what you have node --version # want v22.19.0 or later bwrap --version # Linux and WSL2 only; any version is fine If node is missing or too old, install it from nodejs.org. The Vinci Code installer also checks your Node version and stops with instructions rather than installing something that will not run. Once it runs, connect it to your account with /login — no API key needed. The full reference lives in the Vinci Code documentation.

Last updated on Jul 30, 2026