Run the AI where you want — cloud, local, or your own servers

The Deputy and Posse are provider-agnostic — the SAME features run wherever you point them. The heavy compute lives wherever the model runs, so your workstation only needs a big spec when the model runs ON it.

1. CLOUD (bring your own key) — Anthropic / OpenAI / Gemini. Workstation stays minimal; known secret shapes are masked before anything is sent. Fastest to start, top model quality.

2. LOCAL (Ollama on this workstation) — fully private, works offline, but this box needs ~32 GB + a GPU (see 'System requirements & workstation specs').

3. YOUR OWN INFERENCE SERVER (private) — point the Deputy at a server on your network and every workstation you point at it runs at the minimum spec:

• Ollama: set the Ollama Host to your server, e.g. ollama.lan:11434 (server: OLLAMA_HOST=0.0.0.0:11434).

Data then leaves this machine for that server, so the Deputy shows the off-machine badge and WARNING before a

send, and the Posse won't send it show-step output.

• Any OpenAI-compatible endpoint: pick the OpenAI-compatible provider and set the Base URL to your endpoint —

self-hosted vLLM / LM Studio, Azure OpenAI, or an enterprise model-serving platform (e.g. a Databricks

serving endpoint). Set base URL + key + model.

4. YOUR DATA (MCP) — connect the Deputy to your systems of record over MCP so it reasons over YOUR network data: ServiceNow, Cisco DNAC, NetBox, a CMDB, or a SQL source. Pick one from the Deputy's 'MCP servers ▸ Add from catalog' (it pre-fills the command; you point it at your system and connect), or configure any other MCP server yourself.

Privacy and redaction: known secret shapes (passwords, SNMP/TACACS+/RADIUS keys, tokens, private keys) are masked in the copy sent to the model — every backend is treated the same, cloud or local Ollama — and in saved transcripts, Incidents and exports. You still see the real values in the transcript. It's a pattern match, not a guarantee: a secret in a shape it doesn't recognise can get through, so don't paste one in on purpose.

Workstation spec: cloud AI and any inference you host elsewhere keep the workstation at the MINIMUM spec — only a model running locally on the box needs the 32 GB + GPU. Put ONE GPU server on your network and keep every workstation you point at it at the minimum spec.