The constitutional hook engine for AI coding agents.
A local, deterministic MCP server that intercepts agent actions before they execute and enforces your rules — not as documentation the model can ignore, but as a decision it can't route around.
npm install -g @sunaiva/gate
Enforcement that runs on your machine.
The Gate intercepts AI agent actions before they execute and enforces constitutional rules. 104 rules ship with the package: 35 enforced locally (27 constitutional + 8 recommended-default) and 69 premium rules served via backend on the paid tier.
Fail-safe by default. Local rules enforce fail-closed — errors exit with a structured audit entry, never silently pass.
Opt-in fail-open available via SUNAIVA_GATE_FAIL_OPEN_ON_ERROR=1 for evaluation environments.
The free tier needs no signup, no API key, no onboarding step.
35 local rules evaluate entirely on-device. Works offline.
Every decision is appended to a JSONL log at ~/.sunaiva/audit/audit.jsonl.
From tool call to verdict in under a millisecond.
Free & local: the 35 rules run entirely on your machine. Nothing leaves.
Free for personal and evaluation use under BUSL-1.1. The 35 local rules evaluate on-device and your audit log is plain JSONL on disk. A representative sample of the constitutional categories — not exhaustive.
Deploys, publishes and financial commands require an explicit token before they're allowed to run.
No writes outside the designated workspace. No destructive git operations.
Gold asset paths are enforced read-only — the agent cannot overwrite protected material.
No shell=True. No empty-index commits. Hygiene the model tends to skip.
On paid: premium rule evaluation sends the action to our server (the detection patterns are proprietary); only content-free verdict metadata — tool, decision, rule ID, timestamp — is retained in your dashboard. The action content itself is not stored.
Step up when one machine isn't the whole team.
Annual prepay is 2 months free across paid tiers (10× monthly).
| Tier | Price | What you get | |
|---|---|---|---|
| Open | $0 · no account | 35 local rules (27 constitutional + 8 recommended-default) · no signup · no API key · zero telemetry · JSONL audit log on disk · evaluate offline | Install → |
| Dev | $0 · with account | Everything in Open, plus the hosted dashboard, shadow mode, and premium-rule previews | Sign up → |
| Pro | $49/mo · $490/yr | + 104 rules total (69 premium + 35 local) · semantic analysis · 50 custom rules · shadow/advisory/enforce modes · 90-day audit retention · Ed25519-signed verdicts · all agent integrations (Claude, Cursor, Codex, Windsurf) · observability + API access · email support (48h SLA) | Get it → |
| Team | $199/mo · $1,990/yr · 5 seats incl. | + everything in Pro · 5 seats included (need more? contact us) · shared custom rules · shared audit dashboard · org cert-chain · Slack/webhook alerts · 1-yr audit retention | Get it → |
| Enterprise | from $999/mo | Custom rules · air-gap deployment · SSO / SAML · dedicated SLA · hardware attestation (Patent 04, targeted GA — contact sales for current availability) | Contact sales |
~/.sunaiva/audit/audit.jsonl) and the CLI.
See every verdict. Tune every rule.
Browse the audit log, toggle rules and manage your plan from the hosted dashboard — no terminal required.
Where the Gate sits in the portfolio.
The Gate is built on the same architecture the Sunaiva Triple-Gate Validation MCP (Patent 01 Claim 1 commercial embodiment) uses. The Gate enforces agent actions — Claim 22 of USPTO 64/006,491, autonomous AI agent action validation.
The full conjunctive AND-logic Triple-Gate over AI outputs (Claims 1 + 3 + 5) ships in the Validation MCP, now live. See the patent portfolio →
Install the Gate. Try the architecture.
Free for personal and evaluation use. No account, no API key.
npm install -g @sunaiva/gate