Using NeuraFrame™

NeuraFrame™ runs as a local service with a small command-line interface and a local API. This page covers day-to-day use.

Check status

neuraframe status

Shows the service state, product version, whether the engine is active or in pass-through, and the license state.

Connect a local model

NeuraFrame™ does not include a model. You need your own model server running and reachable over HTTP first, for example a llama.cpp completion server, plus an embedding server for semantic reuse. Point NeuraFrame™ at them, then restart the service:

neuraframe config set completion_url http://127.0.0.1:8080/completion
neuraframe config set embedding_url http://127.0.0.1:8081/embedding
sudo systemctl restart neuraframe-studio

Only safe knobs can be set this way (host, port, and the model endpoints). Everything else lives in /etc/neuraframe-studio/config.json.

Ask a question

neuraframe ask "What is the default power mode?"

The first time, the model answers and the result is remembered. Ask the same thing again and it is served from memory with no model call. The response shows the source (for example model or meaning_memory_exact) so you can see when reuse happened.

Teach a correction

Prompting gets a response. Teaching changes the future response. When an answer is wrong, correct it once and the correction is applied to future matching work instead of disappearing into a chat history:

neuraframe correct "The default power mode is MODE_30W." "The default power mode is MODE_50W."

See what you are saving

neuraframe stats

Reports reuse counters: total requests, model calls made, calls avoided, and memory used. This is where the value shows up over time as recurring work is served from memory rather than recomputed.

Run the example

Prerequisite: a local model must be running and connected first (see Connect a local model, above). The example calls your model, so without one it cannot run. This is the step most people miss on a fresh machine.

neuraframe demo

Runs a bundled example end to end: it ingests a short document, asks a question, then asks again and serves it from memory, with the model called only once. A quick way to confirm the reuse behavior on a new install.

The local API

The same capabilities are available over a local, token-authenticated REST API for integrating NeuraFrame™ into your own workflows (health, ask, correct, stats, ingest). The API listens locally only; nothing is exposed off the device by default.

Reuse is only applied when the system is confident. Uncertain cases still go to the model or are routed for review, and novel inputs always reach the model. NeuraFrame™ reduces repeated work; it does not replace your model.

← Licensing & trial Next: CLI reference →