Stanchion for VS Code
Local-first runtime governance for your AI coding agents — on your machine, fully offline.
🚧 PUBLIC BETA — 30-DAY FREE TRIAL 🚧
This is the public beta of Stanchion, released for open testing.
It runs free for 30 days, after which a license key is required to continue.
More info: stanchionai.com · Support: contact@stanchionai.com
BETA — public testing build. You're running a pre-release sample of
Stanchion. It is feature-complete for single-seat use but still under active
testing; please report issues to contact@stanchionai.com. The trial runs for
30 days from first launch — after that, activate a license key
to keep governance running. Until then everything is unlocked.
Stanchion watches what every AI coding agent and chat surface actually does — tokens, cost, models, tool calls, file writes — and lets you govern it: approve or deny models, cap spend, block risky actions, and keep a tamper-evident audit trail. Everything runs against a local daemon on 127.0.0.1; no code or prompts leave your machine.
This extension is the in-editor window into that: a live session sidebar plus the full InsightsHub dashboard.
Learn more at stanchionai.com · Questions? contact@stanchionai.com
Features
A bespoke, real-time view of the active coding session:
- Token usage & budget burn-down, with projected cost
- Context window % gauge (auto-detects the model's window, promotes to its max)
- Recent prompts and the current model
- Threat / lockdown status — see at a glance when governance has intervened
📊 InsightsHub — the full dashboard
Open the complete operator dashboard right inside the editor (Stanchion: Open InsightsHub):
- Universal Telemetry — every capture source (Claude Code, Copilot, Cursor, Codex, Gemini, Bedrock, and the in-browser chats) in one place
- Cost Management — spend by model, by service, by session
- Security ▸ Models — per-model governance, mirrored across two surfaces:
- IDE / CLI / Proxy controls, and
- Web Interface controls (ChatGPT, Claude.ai, Gemini, Copilot web)
- Argus, Crews (multi-agent), Compliance, and MCP gateway views
🛡️ Model governance (IDE and Web)
For every model, on each surface independently:
- Allow / Deny — denied models return HTTP 403 (IDE/proxy) or are blocked at send-time (web)
- Control toggles — tool calls, file writes, git mutations, network, sub-agent/crew spawning
- Usage caps — per-session and per-day token limits
- Cost caps — per-session and per-day dollar limits
- External — web access, data export
- YAML push — export/import policy per surface, so IT can ship one file to govern a fleet
🌐 Live web send-gate
Deny a model on the Web surface and the companion browser extension blocks the send on chatgpt.com / claude.ai / gemini.google.com — version- and tier-aware (deny "Opus" → all Opus blocked; "GPT-5.5" → only that model).
🔐 Audit, recovery & repair
- Verify Audit Chain — confirm the tamper-evident ledger is intact
- Emergency Rollback — revert to a known-good state
- Repair — auto-fix a stuck daemon or extension wiring
Commands
| Command |
What it does |
Stanchion: Start Daemon / Stop Daemon |
Control the local governance daemon |
Stanchion: Open Insights |
Open the live session sidebar |
Stanchion: Open InsightsHub (Full Dashboard) |
Open the complete operator dashboard |
Stanchion: Break-In (Inject Guidance) |
Inject a steering message into the active agent |
Stanchion: Verify Audit Chain |
Validate the tamper-evident audit ledger |
Stanchion: Emergency Rollback |
Roll back to a known-good state |
Stanchion: Repair |
Auto-fix the daemon + extension |
Stanchion: Reload Auth Token / Refresh All Views |
Re-auth / refresh the UI |
Settings
| Setting |
Description |
stanchion.daemon.url |
Daemon URL (default http://127.0.0.1:52737) |
stanchion.daemon.token |
Bearer token for authentication |
stanchion.contextWindow |
Context-% gauge window (auto / 200k / 1m) |
stanchion.sidebar.mode |
native (live dashboard) or unified (embedded Insights SPA) |
stanchion.captureThirdPartyAssistants |
Extend metadata capture to additional third-party assistants (off by default) |
Captured assistants
First-party out of the box: Copilot, Continue, Cody, Tabnine, Amazon Q, Cline, Cursor, Windsurf, plus Claude Code, Codex, Gemini and the in-browser chats. Optional (captureThirdParty): Augment, Refact, Supermaven, standalone Codeium, Qodo, IntelliCode, GitHub Models. Stanchion records only metadata — kind, language, file extension, text length — never suggestion content.
Installation — just click Install
That’s it. This is a self-contained extension: it ships the Stanchion daemon
inside it and starts it for you. There is nothing else to download and no admin
rights required.
On first activation the extension sets up and launches the local daemon
automatically (you’ll see a brief “Setting up the local daemon…” notification).
Within a few seconds the status bar reads Stanchion · OK and the sidebar goes
live. Everything runs on 127.0.0.1 — no code or prompts ever leave your machine.
- Works on macOS and Windows out of the box — the Marketplace serves the build
matching your platform.
- No separate installer, no service setup, no PATH wrangling. The bundled
daemon self-starts and survives this window closing.
- If it ever shows offline, run Stanchion: Repair from the Command Palette.
Beta trial & licensing
This public beta is free for 30 days from first launch. When the trial ends,
Stanchion asks for a license key to keep governing — grab one or read the details
at stanchionai.com. Need help or have feedback?
Email contact@stanchionai.com.
Advanced / fleet deployments (optional): prefer a machine-wide daemon managed
as a Windows Service or shared across non–VS Code surfaces (CLIs, the browser
extension)? A standalone installer is available at
https://github.com/dw425/Stanchion-v2.3-Beta/releases/latest. It’s not
required — the extension is fully self-contained on its own.