Off Grid Console is the control plane for agentic AI. One governed gateway every model call, agent action, and byte of data passes through, running on your own infrastructure, fully auditable, built on open source. Manage a fleet of on-device nodes, ground them in your knowledge, and prove compliance to a regulator.
Your people are already using AI. The question is whether you can see it.
Today model calls leave through a dozen tools, on data you cannot track, at a cost no one owns. Console puts one chokepoint in front of all of it. Every call is routed, observed, costed, and kill-switchable. Nothing runs that you did not allow, and everything that runs is on a record you can defend.
What you get, in one console
One AI gateway
Every model call goes through a single endpoint you can route, observe, cost, and kill-switch. One chokepoint, every provider, on-prem or cloud. This is what ends shadow AI.
Guardrails and policy
PII and prompt-injection screening on every request, plus role and attribute access control over who can use which model, data, and tool. The policy is set once and enforced everywhere.
Your knowledge, grounded
A private RAG brain over your own docs and systems, with verified citations, so answers are grounded in your context, not guessed. Off Grid distills how work actually happens into one brain every agent and person draws from.
Agent QA
Offline evals, live LLM-as-judge scoring on real traffic, and drift detection. One surface that answers whether the agents are still doing a good job, and if not, which one regressed and when.
Provenance and tamper-evidence
Every report export carries a signed, offline-verifiable manifest. Images get C2PA content credentials, artifacts can be Sigstore-attested. Prove what was produced, by whom, unaltered, with only a public key.
Fleet Control
MDM for AI, built on open source (FleetDM and osquery). Provision, govern, and observe every AI-enabled device from one console. Push policy down, pull audit up, kill-switch on demand. Your workforce runs AI, you keep control.
Audit and report packs
An append-only record of every model call, tool call, and byte of egress. Report packs a regulator can defend, mapped to IRDAI, RBI, SEBI, and DPDP. Compliance is a query, not a fire drill.
FinOps
Per-user and per-project cost, virtual keys with budgets, and chargeback. You see what AI costs the business, by team, before the bill arrives. No per-token fee to us, ever.
Five layers, one plane
Layer
What it does for you
Data
Connect, ingest, and mask PII into AI-ready data on your own infrastructure
AI
Model serving through the gateway, the knowledge brain, grounding, and caching
Control
The gateway chokepoint: guardrails, policy, identity, audit, and kill-switch
Regulatory
Report packs, the governance registry, and data residency
Consumption
The console, Fleet Control, FinOps, agents, and reports
Why it is different
You own it
Self-hosted and on-prem. Models run on your infrastructure, data stays in your control. Nothing routes through a server we own. The privacy guarantee is architectural, not a policy promise.
No lock-in
Every capability is a port with an open-source default. Swap any underlying tool, the model server, the vector store, the gateway, with one environment variable. Take the whole plane or just one part.
No vendor tax
All open source under AGPL. You do not pay per token, per key, or per seat. The economics scale with your hardware, not with a meter someone else controls.
Buy only what you want
Nothing is all or nothing. Every capability is API-first and adoptable on its own.
Just the API. The gateway, brain, and agents as headless services with documented APIs.
API plus Console. This UI over any subset of those services.
Just the Brain. The ingestion to retrieval (RAG) pipeline, standalone.
The whole control plane. Every layer, unified and governed.
Where Console fits
Off Grid is one private AI stack, from the phone in a pocket to the org’s whole fleet. The consumer apps put it in millions of hands and make it a household name. The Console is where it runs a company.
Same principles at every scale: the model runs on hardware you control, the data stays yours, and the code is open so you can check. See the consumer suite: Mobile, Desktop, and Sync.
