Apps · real-world and reference

Apps

The Lab measures apps that issue requests to a local model. Each app has a kind — reference, a small controlled app, or real-world, a live product — and a status from its catalogue: live when it can produce a run today, planned until then. Each scenario is a named request pattern; a recorded scenario has a committed bundle.

Finboard live

Personal Finance App · Local assistant

kind real-world

Finboard is a local-first portfolio look-through desktop app with a local LLM assistant over typed, read-only capabilities. It exercises tool selection, multi-turn context, narration, privacy and confinement, the runtime's lifecycle and inference performance. Finboard keeps its own eval harness and ledger; the Lab consumes what Finboard exports and links the rest.

scenario	input	status	runs
`runtime-direct`	`synthetic`	recorded	runs
One streamed request sent by the collector directly to the runtime Finboard ships, on loopback, with a synthetic prompt and the runtime's default system prompt. It exercises the runtime the app ships; it does not exercise Finboard's prompt assembly, capabilities or narration.
`assistant-turn`	`real`	planned
`narration`	`real`	planned
`multiturn`	`real`	planned

catalogue · On GitHub ↗

Finboard demo: cycling through Holdings, Analysis, Performance, Rebalance and Retirement on the synthetic demo portfolio — Finboard on its synthetic demo portfolio. Open the demo

Media Search Agent planned

Media search app · AI search

kind real-world

Media Search Agent would bring a second domain — retrieval, ranking, multimodal context and orchestration over local data — once its generative integration exists. Its scenarios are defined with its first bundle; until then nothing is measured.

catalogue · On GitHub ↗

Media Search Agent demo: search, browse, people and video — Media Search Agent itself: search, browse, people and video. Its integration with the Lab is planned, so nothing here is measured. Open the demo

LLM Probe planned

Command-line test app · Controlled test runs

kind reference

LLM Probe will be a small, deterministic command-line app for exploring the model and runtime stack without a domain: change one variable, observe what happens. It will be the control, not a substitute for a real-world app, and every byte of its prompts will be authored for the Lab.

scenario	input	status
`completion`	`synthetic`	planned
`context-scaling`	`synthetic`	planned
`tool-call`	`synthetic`	planned
`long-generation`	`synthetic`	planned