1f1flowbase
Use cases

Practical model workflows for local AI agents.

Start from a concrete limitation in your current agent setup, then publish the improved behavior as one reusable virtual model.

01

Mount vision for text-first coding models

The limitation

Text-first coding models cannot directly understand screenshots, user interfaces, charts, or PDF pages, leaving the planner without critical context.

The workflow

Expose a vision model or OCR engine as a callable tool. The main model keeps planning and invokes multimodal intelligence only when visual evidence is needed.

Read the implementation guide →
01Claude Code
↓
02Virtual model endpoint
↓
03Main text model
↓
04Vision tool
↓
05Final answer
02

Publish a multi-model review panel as one model

The limitation

A single model has blind spots, but making every client manage parallel calls, multiple providers, and synthesis creates repeated orchestration work.

The workflow

Use a Fusion workflow to ask several reviewers in parallel and synthesize one result. The client still sees a single model name.

Read the implementation guide →
01Agent request
↓
02Parallel reviewers A/B/C
↓
03Synthesis model
↓
04Virtual model response
03

Publish an observable model API

The limitation

Traditional model proxies often expose one HTTP request but cannot explain the models and tools behind the answer or where the cost actually occurred.

The workflow

Publish the workflow itself through an OpenAI- or Claude-compatible API and preserve the connected nodes, calls, tools, tokens, duration, and errors.

01Standard model API
↓
02Workflow runtime
↓
03Execution trace
↓
04Cost and failure attribution