Frequently asked questions

vigie is a local-first monitoring daemon for Claude Code. It watches every file change, every command, every dollar spent. It catches loops, scope drift, and context bloat before they waste your time. Checkpoints let you roll back before anything ships.

Run vigie start in your project. A lightweight daemon spawns Claude Code, captures every event — file edits, shell commands, token usage — and streams them to a real-time dashboard. No code leaves your machine.

The daemon runs entirely on your machine. It observes session metadata — files touched, commands run, token counts — but never reads or stores your source code. The dashboard connects via a local websocket and just streams command output — your codebase stays on your machine.

vigie is currently in private beta. Join the waitlist to get early access — you’ll receive install instructions as soon as the beta opens. Installation is a single command.

Claude Code at launch. OpenCode (which supports Copilot licenses), Codex CLI, Aider, and Cline are on the roadmap. The architecture is CLI-agnostic by design — if your tool has a CLI, vigie can monitor it.

macOS or Linux (Windows via WSL). Git installed. Claude Code as your AI CLI. The daemon is lightweight and runs in the background with minimal CPU and memory overhead.

Every file read, created, or edited. Every shell command. Token usage, context window percentage, and running cost. All streamed live to your dashboard as events happen.

vigie detects edit → test → fail → edit cycles where the agent is stuck repeating the same pattern. It flags the loop in your dashboard and suggests a prompt to help the agent break out.

When your agent starts touching files outside the scope you declared for the task, vigie flags it immediately. You see exactly which files were accessed outside scope so you can course-correct before the agent goes further off track.

A live gauge shows how much of the context window is consumed. At 65%+ vigie warns that the agent may start losing early instructions and suggests opening a fresh window from your last checkpoint.

Checkpoints are approval gates that snapshot your project state. If something goes wrong, you roll back to the last clean state and resume with just the necessary context — no reprompting needed.

Yes. vigie supports monitoring multiple concurrent agents, each with its own activity stream in the dashboard.

A live cost counter tracks spending in real time. vigie hints when a cheaper model would suffice for the current task. You can set budget guardrails that pause the session before exceeding a threshold.

Yes. vigie is open source under the MIT license. All features — monitoring, loop detection, scope drift, checkpoints, rollback, cost guardrails — are free and always will be. No paywall, no premium tier.

vigie never touches your AI tokens. You pay Claude (or any supported provider) directly at their standard rates. vigie is a monitoring layer — there’s no hidden per-token fee and no vendor lock-in.

In the agent era, every license is MIT. Agents clone, fork, and ship code faster than any compliance team can review headers. We chose the license that matches how software is actually built today — no friction, no gotchas.

Yes. MIT license means you can use, modify, and distribute vigie freely — commercially or otherwise. No CLA, no contributor agreement required.

vigie is built by Tmonier SRL, a freelance software consultancy founded by Damien Meur. The same people who steer the ship also keep watch.

No. vigie observes session metadata only — file-change counts, token usage, context window percentage, session duration. It never reads, parses, or stores your source code.

Token counts, context window utilisation percentages, session durations, file-change counts (not contents), and checkpoint events. Never file contents, diffs, prompts, or AI responses.

Yes. vigie is local-first — your data stays on your machine. For hosted dashboard features, all data is processed in EU-based infrastructure.

Yes — it’s open source. Clone the repo, deploy on your own infrastructure. Full control over data residency, access policies, and integrations.

Ollama support is on our roadmap. Once available, you’ll be able to run fully private coding sessions where everything stays on your machine.

Claude Code gives you raw agent output. vigie adds a real-time dashboard, loop detection, scope drift alerts, context window monitoring, cost tracking, checkpoints & rollback, and suggested prompts to unblock your agent. It’s the visibility layer that makes Claude Code predictable.

No. vigie is an additive monitoring layer that runs alongside your existing tools. You keep using your IDE, terminal, and Claude Code exactly as before. vigie just watches and alerts.