Suveren Gateway
HAP is the protocol. Suveren is an implementation of it.
The Human Agency Protocol (HAP) is the open standard for bounded AI-agent authority — it defines the roles (Authority Server, Gatekeeper, Executor) and the concepts (profiles, gates, attestations, bounds, context, receipts). Suveren implements them: this repo is the Gateway, and
suveren-asis the Authority Server. The protocol is open — anyone can build their own compliant gateway.
This repository is the Suveren Gateway — Suveren's implementation of the HAP
Gatekeeper + Executor roles. It runs locally and verifies every tool call
against its authorization before the call reaches an external service. (The
@hap/core library inside this repo keeps its HAP name because it re-exports the
open protocol library — "HAP-compliant" and spec references describe the open
standard, not the Suveren brand.)
Part of suveren.ai.
Let your AI agents act — within bounds you control.
The gateway runs on your machine, between your AI agents and the tools they use — payments, email, CRM, deployments, infrastructure. Your agents go through a local policy layer before reaching external services. Nothing executes without authorization.
Works with any MCP-compatible agent. Define and authorize what they're allowed to do. Every action is bounded, time-limited, and traceable to a human decision — so agents can execute safely at scale.
Automatic or Review
Set the threshold. Routine actions execute automatically within the bounds you defined. High-stakes actions pause for your review before the agent acts.
Automatic — You commit to specific bounds upfront: max amounts, allowed actions, time windows. The agent executes autonomously within those bounds. For each tool call the gateway verifies your authorization and requests a receipt from the Authority Server, which issues the signed receipt before the call runs — no receipt, no execution.
Review each action — You define bounds but defer full commitment. When the agent proposes an action, you review it in the gateway UI — seeing exactly which tool, which arguments, which context. You approve or reject. Execution only proceeds after your decision.
Both modes are bounded. In both, the Authority Server issues a signed receipt before the action runs — no receipt, no execution — and that signed history is a full audit trail. The difference is whether you trust the bounds enough for autonomous execution, or want to review each action individually.
How It Works
Human AI Agent
| |
| 1. Define bounds, |
| articulate direction, |
| commit (or defer) |
v |
Authority Server |
| 2. Sign attestation (Ed25519) |
v |
Gateway |
| 3. Connect via MCP ----->|
| 4. Tool call <-----------|
| |
| 5. Fully committed: |
| Gatekeeper checks bounds, asks AS |
| -> AS issues receipt, execute |
| |
| Deferred commitment: |
| -> proposal created |
| -> human reviews in UI |
| -> commit or reject |
| -> on commit: AS receipt, execute |
The agent never holds credentials or signing authority. It acts within the bounds you set — high autonomy without losing accountability.
What the Agent Sees
When an agent connects, it receives a compact authority brief — active authorizations with bounds, live consumption, and available tools:
=== ACTIVE AUTHORITIES ===
[spend-routine] [email protected] (45 min remaining)
Bounds: amount_max: 100, currency: USD, action_type: charge
Usage: $234/$500 daily, $1280/$5000 monthly, 8/20 tx
Intent: Enable automated purchasing for business operations.
4 gated tools, 19 read-only
No credentials. No signing keys. Just the scope of what the agent is allowed to do — and the human's stated reason for granting it.
Quick Start
Pick whichever fits — all three produce the same gateway. For company laptops managed by IT, use Option C.
Option A — Docker
Requires Docker.
docker run -d --name suveren-gateway \
-p 7400:3000 -p 7430:3030 \
-v $HOME/.suveren:/app/data \
ghcr.io/suverenai/suveren-gateway
Open http://localhost:7400. The MCP server is at http://localhost:7430.
Option B — npm
Requires Node.js 22+.
npm install -g @suveren/gateway
suveren-gateway start # runs in foreground; Ctrl+C stops
# or
suveren-gateway start --detach # runs in the background; data + logs in ~/.suveren/
suveren-gateway status # check it's up
suveren-gateway stop # stop a detached run
Open http://localhost:3400. The MCP server is at http://localhost:3430.
To upgrade later, run these two commands:
npm install -g @suveren/gateway@latest
suveren-gateway restart
(Written as two lines on purpose: && is not a valid separator in Windows
PowerShell 5.1, which is what ships with Windows. Use ; there if you want
them on one line.)
Option C — Windows installer (company laptops)
For laptops managed by a company's IT: no admin rights, no Node.js and no access to the public npm registry needed. Every GitHub Release from 0.13.0 on carries:
The installer is not signed by us. IT checks where it comes from (sha256sum -c SHA256SUMS and gh attestation verify <file> -R suverenai/suveren-gateway, which shows the exact commit it was built from), signs it, and distributes it, e.g. with Intune. IT can preset the Authority Server, simulation mode, proxy and certificate centrally; the employee cannot change them, and updates come from IT. Step by step: Windows IT guide · settings: Managed settings.
Open http://localhost:3400 (Start menu: Suveren Gateway). The MCP server is at http://localhost:3430. Other ports or another data folder: msiexec /i suveren-gateway.msi PORT=3500 MCP_PORT=3530 DATA_DIR="D:\Suveren", or the "Gateway settings" page of a double-click install — see the IT guide.
Ports and data folder
npm and the Windows installer default to port 3400 (app), 3430 (AI assistants) and the data folder ~/.suveren. To change them for good — kept across restarts, upgrades and autostart:
suveren-gateway stop # if you started it with --detach
suveren-gateway config set port 3500
suveren-gateway config set mcp-port 3530
suveren-gateway config set data-dir /Volumes/Data/suveren
suveren-gateway config get # shows each value and where it comes from
suveren-gateway config unset port # back to the default
Changing the data folder does not move existing data — copy the old folder yourself first if you want to keep it. After changing mcp-port, update the MCP address in every AI assistant you connected. Environment variables (SUVEREN_CP_PORT, SUVEREN_MCP_PORT, SUVEREN_DATA_DIR) still win over saved values; IT policy wins over both (Managed settings).
Connecting an MCP client
Every option exposes the same MCP transports — use the port from the option you chose (7430 for Docker, 3430 for npm and the Windows installer):
Streamable HTTP: POST http://localhost:<port>/mcp
SSE transport: GET http://localhost:<port>/sse
Local development
Running from source gives you hot-reload across all three services:
cd suveren-gateway
pnpm install
pnpm dev # UI on :3400, control plane on :3402, MCP on :3430
See docs/development.md for environment variables, testing, and per-service dev commands.
Pinning the Authority Server's TLS certificate
Opt-in, for self-hosted Authority Servers with a stable signing key — not needed against the hosted suveren.ai, and off by default.
The gateway already verifies the Authority Server can sign under the right key before it ever sends an API key or session cookie. What that alone does not cover: something sitting on the network path between the gateway and the Authority Server, presenting its own TLS certificate, that a locally-trusted CA (e.g. a custom --ca-file) would otherwise accept. TLS certificate pinning closes that gap by pinning the Authority Server's certificate public key (SPKI, SHA-256) — every connection after that must present the same key, or the gateway refuses it and locks.
Pinning only protects from the moment the fingerprint has actually been checked over a second channel — a phone call, a video call, a separate trusted connection. That is why enabling it requires --expect-fingerprint: not an optional confirmation step, but the check itself. Recommended: enable it once, right after pairing, from a network you already trust.
# Get the Authority Server's own fingerprint independently — ask the operator,
# or run this yourself on a trusted network:
openssl s_client -connect your-as-host:443 </dev/null 2>/dev/null \
| openssl x509 -pubkey -noout \
| openssl pkey -pubin -outform der \
| openssl dgst -sha256
# Compare it against what you were told out-of-band, THEN enable:
suveren-gateway config set pin-tls on --expect-fingerprint <the-sha256-you-just-confirmed>
suveren-gateway restart
A few things worth knowing before turning it on:
- Certificate renewal with the SAME key keeps the pin working — nothing to do. Renew with the same key where your tooling supports it (e.g.
certbot renew --reuse-key). - A renewal under a NEW key locks the gateway (
as-tls-mismatch) until an operator re-pairs — this is deliberate fail-closed behavior, not a bug. Re-pairing means clearing<dataDir>/as-pairing.jsonand signing in again. - Pinned connections trust Node's own bundled root certificates plus
--ca-file/NODE_EXTRA_CA_CERTS— not your operating system's trust store. suveren-gateway config set pin-tls offdisables enforcement at any time; the stored fingerprint stays on file, so turning it back on later with the SAME--expect-fingerprintvalue succeeds immediately (it is still required every time — stating it again is cheap, and the command refuses outright if it doesn't match what's on file).
See suveren-gateway config help for the full command reference.
Behind a company proxy
On a company laptop that routes all internet traffic through an HTTP(S) proxy, the gateway needs two things: the proxy address, and — if that proxy does TLS inspection — the company's root certificate.
The proxy. Set the standard environment variables (upper or lower case both work) before starting the gateway, or save one with the CLI:
export HTTPS_PROXY=http://proxy.corp.example:8080
export HTTP_PROXY=http://proxy.corp.example:8080 # usually the same value
export NO_PROXY=internal.example.com # optional: hosts to reach directly
suveren-gateway start
# or, to save it so every future start picks it up without re-exporting:
suveren-gateway config set proxy http://proxy.corp.example:8080
suveren-gateway restart
Every outbound call this applies to — the Authority Server, the update checker, and a remote AI assistant endpoint — honours the proxy. A loopback target (the control plane and MCP server talking to each other, or a local AI assistant such as Ollama) is never proxied, no matter what is set — a corporate proxy has no route back to the machine it's running on.
TLS inspection. Many corporate proxies intercept HTTPS by re-signing every certificate with their own root — that includes the Authority Server's. If you see a certificate error mentioning an untrusted issuer, trust the company's root the same way you would for a self-hosted Authority Server:
suveren-gateway config set ca-file /path/to/company-root-ca.pem
suveren-gateway restart
This works through the proxy exactly like it works directly — --ca-file applies to the certificate the gateway actually sees, proxied or not.
The one combination that still refuses, on purpose: TLS pinning (opt-in, off by default) together with a TLS-inspecting proxy. Pinning checks the certificate the gateway sees through the proxy — if that's the proxy's own re-signed one rather than the real Authority Server's, the pin will not match, and the gateway refuses the connection and says so, even with the company root trusted via --ca-file. That refusal is correct: ask network/IT to exempt the Authority Server's host from TLS inspection, the same request you'd make for any pinned certificate on that network.
On a company-managed machine, IT can set the proxy centrally instead of relying on HTTPS_PROXY being exported in every shell — see Managed settings. A policy-set Proxy (Windows registry or a JSON policy file) overrides an already-set HTTP_PROXY/HTTPS_PROXY rather than merely filling it in, and config set proxy / --proxy refuse with "set by your IT" when it's locked. A loopback target is still never proxied, regardless.
Simulation mode — block every real system
A mandate is bound to a profile (e.g. "sales"), not to a specific connector. If a real connector (e.g. a live ERP account) and a simulated one (the built-in ERP/CRM/email simulators) share a profile on one gateway, a mandate meant only for testing also authorizes the real connector — nothing about the mandate says which one it's for.
Simulation mode closes that gap gateway-wide: turn it on and every connector without a
manifest simulation marker is refused to even start (the built-in ERP, CRM, and email
simulators all declare one; a connected Gmail or live ERP account does not), and every
connector that does declare one has its mode forced to simulation, overriding whatever
credential value is on file. Mandates and profiles are completely unaffected, and the
Authority Server never learns this is on — it's a purely local, gateway-side switch.
# Start fresh with real systems blocked:
suveren-gateway start --simulation
# Or flip it on an existing install (takes effect on the next start/restart):
suveren-gateway simulation on
suveren-gateway restart
# Check the mode (the SAVED setting and, if the gateway is running, the LIVE one):
suveren-gateway simulation status
# Turning it off makes real systems reachable again, so it requires typed
# confirmation (or --confirm live for scripts) — turning it ON does not:
suveren-gateway simulation off
suveren-gateway restart
suveren-gateway status also shows the current mode. See suveren-gateway simulation help
for the full command reference.
Technical Documentation
Protocol specification: humanagencyprotocol.org
License
MIT — see LICENSE.
