Documentation

Start with the control path.

Connect an existing AI client, verify policy before provider egress, and keep the operational boundary explicit from the first request.

Documentation setHormuz v0.1 alpha
Source-linked
Evidence-bounded
Open gates visible

Run the control plane locally.

The local path proves routing and policy behavior. It is not a production deployment profile and should not be exposed directly to the public internet.

TerminalPython 3.11+
cp config.example.json hormuz.json
export HORMUZ_TOKEN="replace-with-a-long-random-token"
export OPENAI_API_KEY="your-company-openai-key"
export ANTHROPIC_API_KEY="your-company-anthropic-key"

python3 -m hormuz --config hormuz.json doctor
python3 -m hormuz --config hormuz.json serve

Credential boundary: employees authenticate to Hormuz; provider credentials stay on the Hormuz server and are never forwarded back to the client.

Four contracts, one request path.

01

Identity

Resolve the organization, team, person, client, and authorization profile before durable access or provider work.

02

Policy

Combine organization, team, and person rules monotonically so lower scopes can tighten controls, never weaken them.

03

Egress

Apply model access, budgets, privacy rules, DLP, and any exact-request approval before the provider call.

04

Evidence

Record bounded identity, policy, usage, cost, and security outcomes without retaining the prompt or response body.

Go deeper by control surface.

Next

Connect an existing AI client.

View integrations