F FanOn

Remove AI logistics from software development

Build software. Let FanOn decide where AI runs.

Local when it is a fit. Provider when the request or availability calls for it. FanOn keeps developers in their existing AI workflow while making execution decisions visible. Reduce unnecessary provider usage without changing how developers work.

Get Started Documentation
Developer-first Privacy-first Transparent routing Local/dev MVP
Developer Cursor / Continue / Cline
FanOn Decides where work runs
Local When it fits routine, private, close
Provider When called for hard, unavailable, specialized

The fan story

The moment is simple: the fan turns on.

Developers have powerful computers in front of them. Most AI requests still go to the cloud by default.

FanOn exists for the moment when local capacity should help, provider calls should be intentional, and the developer should not have to think about any of it.

Problem / Observation / Belief

AI moved into the IDE. The execution decision did not.

Coding assistants spread fast. Local models became useful. Provider costs and privacy questions became real. But the default path still sends almost everything outward.

Problem Every tool becomes another place to configure models, keys, cost, and fallback.
Observation Many everyday development tasks do not need a frontier cloud model.
Belief Use local when it is a fit. Use providers when the request or availability calls for it.

Why teams adopt FanOn

Practical outcomes without making developers manage the plumbing.

FanOn is not a generic cost dashboard. It is a way to make better AI execution decisions while preserving the workflow developers already use.

Less waste Reduce unnecessary provider usage.
More privacy Keep more eligible work local or on team-owned capacity.
Better use Put existing developer hardware to work when it is a fit.
More clarity Make AI execution decisions visible instead of implicit.
Same workflow Keep developers in the AI tools they already use.

How FanOn works

A small decision layer between developer tools and AI execution.

Today, FanOn supports two validated integration modes. Gateway Mode lets an IDE or tool talk to FanOn directly. Agent Tool Mode lets an AI coding assistant use FanOn for bounded local subtasks. In both modes, FanOn helps choose where suitable work runs and makes execution decisions visible.

1 Gateway Mode

An IDE or tool talks to FanOn directly and keeps its familiar workflow.

2 Agent Tool Mode

An AI coding assistant calls FanOn for bounded local subtasks.

3 Visible decisions

FanOn chooses where suitable work runs and shows the route it took.

Why FanOn

Local AI needs a product layer, not another pile of configuration.

Ollama runs local models. MCP connects tools. FanOn turns those pieces into developer-facing capabilities, readiness checks, policies, metrics, and setup instructions that coding assistants can use without making every developer become an AI infrastructure operator.

Not just Ollama FanOn organizes local models into capabilities, diagnostics, and workflows.
Not just MCP MCP is the protocol; FanOn is the local execution and capability layer.
Local first Eligible tasks run close to the developer before reaching for cloud providers.
Visible Readiness, usage, and dashboard output make execution decisions inspectable.

Trust

Useful routing has to be explainable.

FanOn is developer infrastructure, not employee monitoring. The product is shaped around aggregate infrastructure signals, visible route reasons, and privacy-first defaults.

Read the trust model

Pilot / status / fit

Useful enough to test. Early enough to shape.

FanOn is a local/dev MVP, actively dogfooded, and not production-ready yet. It is currently being dogfooded through real Codex, FanOn CLI, and local-model workflows. We are looking for design partners dealing with AI provider sprawl, rising AI costs, local model experimentation, or privacy concerns around cloud-by-default AI workflows.

Status Local/dev MVP in learning and validation.
Best fit Engineering managers, staff engineers, platform teams, and AI infrastructure teams.
Workflows Cursor, Continue, Cline, local model experiments, and AI cost optimization.

Design partners

Help shape the layer that should already exist.

We are looking for conversations and feedback from teams who want AI execution to feel simpler, more private, and more intentional.

Join the Design Partner Program

Takes about 2 minutes. This is a design-partner conversation, not a sales process.