Traffic history
Local overview of observed traffic history inside the self-hosted environment.
Skilful Fox
Independent research studio for applied AI systems
GreyFox
Controlled AI traffic tooling for local operator workflows.
GreyFox intercepts and routes local AI application traffic, allowing you to inspect, debug, and enforce policies on LLM data streams. It keeps the control plane self-hosted and telemetry-free, without sending operational data to the cloud.
Interface
A local operator panel, not a hosted surveillance layer
GreyFox is designed as a self-hosted control surface for practical AI traffic workflows. The UI exists to help operators reason about local routing, users, providers, and usage signals without sending operational data outside the container boundary.
More interface surfaces
How it works
Route only what you choose through GreyFox
GreyFox sits between your application and an OpenAI-compatible provider only when you point traffic at its local proxy endpoint. The application still decides when to call AI.
Local proxy integration
Change the base URL, keep the rest of your client flow.
GreyFox runs as a local proxy server. Instead of sending requests directly to an OpenAI-compatible provider, point your LLM client at the GreyFox endpoint and keep control inside your own Docker environment.
- Application code
- Your app still decides when to call the model and which requests should be routed.
- GreyFox boundary
- GreyFox observes only traffic explicitly sent to its local proxy URL.
- Provider connection
- The proxy forwards compatible requests onward after local checks, limits, and logging.
TypeScript
Base URL swap// Before: direct API call
// const openai = new OpenAI({
// baseUrl: 'https://api.openai.com/v1'
// });
// After: routed through GreyFox for local control
const openai = new OpenAI({
baseUrl: 'http://localhost:XXXX/v1'
}); Use your actual GreyFox host and port in place of localhost:XXXX. The same pattern applies to any OpenAI-compatible client that supports a configurable base URL.
Install from the command line
The package is published through GitHub Container Registry.
$ docker pull ghcr.io/skillful-fox-studio/grey-fox-community:0.1.1GreyFox only sees requests that are explicitly routed through its proxy endpoint. No hosted GreyFox control plane is involved, and no telemetry is sent outside your self-hosted Docker environment.
Editions
Community, Essential, and Pro
GreyFox now has a public Community Edition and a released Essential subscription. Pro is in development for deeper policy, audit, and quota-control workflows.
Community
Open sourceA public, telemetry-free baseline for inspecting local AI traffic and controlling simple proxy behavior inside your own environment. Version 0.1.1 is a maintenance patch focused on UI template and styling refinements.
Community Edition Scope
- Local proxy inspection.
- Zero telemetry or hosted control plane.
- Manual route overrides.
- Single-profile storage.
Essential
Commercial editionA released paid edition for repeatable local AI routing setups, richer request shaping, and deep traffic analytics views.
Included in Essential 0.2.1
- Community Edition baseline plus commercial workflow features.
- Up to 25 active managed users by default, controlled by license metadata.
- Up to 5 saved environment profiles.
- Up to 10 deterministic request/response presets.
- Advanced analytics workspace for tokens, cache, traffic sources, models, guards, and latency.
Repository access is provided by private GitHub invite after purchase.
Pro
Advanced edition
An advanced edition in development for audit-heavy local AI traffic workflows, programmable routing, and shared policy artifacts.
Development scope
- Enterprise audit log exports.
- Custom routing policy scripts (JS/WASM).
- Configurable limit and quota schedules.
Not available yet
Design principles
Operator-first, not telemetry-first
- Telemetry-free by default: operator tools should not quietly become observation tools.
- Local-first orientation: keep useful AI workflow control close to the machine and the developer.
- Clear commercial boundary: paid editions should fund development without weakening the open Community Edition.
Workflow direction
From local experiments to repeatable control
- Inspect and reason about AI traffic paths during local experiments.
- Create a controlled surface around proxy flows instead of scattering configuration across scripts.
- Bridge research utilities and productized operator workflows without pretending to be generic enterprise software.