Ari’s Space*   est. 2026
Skip to content
ari*s space[ a. ]
← Libraryari os / 2026-06-15

Advanced: choose what runs

Inspect the optional memory capabilities, their configured defaults and the distinct requirements for LENS cards and image captioning.

7 parts / Hover to explore. Click to read.

Inspect the optional memory capabilities, their configured defaults and the distinct requirements for LENS cards and image captioning.

[LENS enabled]
  +-> card lookup -> inspect card
  |
  +-> LLM enabled
        + working vision backend
        -> image caption
        -> inspect result
Card lookup does not need a vision-model call. Seeing an image additionally requires the enabled LLM setting and a working vision backend in the checked source.
Context Pack (for agents)

The core brain doesn't need every optional capability switched on. Start with the work you need it to do, inspect the current settings and enable one additional surface at a time.

This page originally called the whole advanced surface default-off. The public configuration checked for this revision is more specific: knowledge-graph extraction, EARS and LENS default off; wander and prediction default on when no saved or environment setting overrides them. Inspect your installed version instead of relying on the old blanket claim.

  • Knowledge graphdefault offExtract entities and relations, then inspect the resulting graph.
  • Predictioncheck configurationThe checked source defaults predictive work on. A command still needs a selected action.
  • Prefetchuses prediction gateRetrieve context for a working directory before its task begins.
  • Mind-wandercheck configurationThe checked source defaults wander on; saved settings and modes affect use.
  • EARSdefault offPrint a local audio transcription or a supported video transcript.
  • LENSdefault offRead a local aesthetic card; image captioning separately needs a working vision backend.

The operations run through python3 -m ari_os.tools.cortex. The control panel is python3 -m ari_os.tools.arios; use its cortex status view to inspect saved settings. Defaults and environment overrides are defined in the checked configuration source.

Extract a graph you can inspect

The knowledge graph turns memory content into entities and relationships. It gives you another structure to inspect alongside the original notes.

A knowledge graph extracted from memory
Extract entities and relationships, then inspect what the graph actually contains.
python3 -m ari_os.tools.cortex kg extract
python3 -m ari_os.tools.cortex kg list
python3 -m ari_os.tools.cortex kg stats

extract requests the extraction pass, list shows entities and stats reports entity, relation and chunk-link counts. Extraction remains gated by the knowledge-graph setting and its model requirements; list and stats are readers. A command being present doesn't mean its backend is ready.

Choose the predictive action

Prediction can prepare material around recurring work. Prefetch retrieves context for a directory before you begin its task.

python3 -m ari_os.tools.cortex predict
python3 -m ari_os.tools.cortex prefetch --cwd <dir>

In the checked CLI, bare predict reports that no action was selected. Use its documented --cluster, --signals or --cwd option for the action you want. prefetch --cwd <dir> names its target explicitly. Both respect the predictive setting. The CLI source is the reference for those choices.

Inspect the returned context before treating it as useful preparation. A repeated pattern is a reason to look, not proof that the next task needs the same answer.

Give wander a focus

Wander looks for an associative connection beyond the material already in front of you.

python3 -m ari_os.tools.cortex wander --focus "<text>"

The focus text supplies the starting point. You can change the feature's saved state through the control panel:

python3 -m ari_os.tools.arios cortex wander on|off

The checked configuration defaults wander on unless an override changes it. The core memory remains usable when you switch it off.

Get a transcript through EARS

The checked EARS command transcribes local audio or fetches a supported YouTube transcript, then prints the text. It does not store that text in memory. Review the result and use the separate memory write path if you intend to retain it.

python3 -m ari_os.tools.cortex ears <file|url>
python3 -m ari_os.tools.arios cortex ears on

Plain audio and transcript work do not require a vision model. They still have their own media/backend requirements. In the checked CLI, EARS refuses work when its toggle or the Cortex LLM setting is off and reports an unavailable transcription backend instead of inventing a transcript.

Separate a LENS card from seeing an image

The LENS command retrieves an aesthetic card by slug.

python3 -m ari_os.tools.cortex lens <slug>

Enable the LENS surface through the control panel:

python3 -m ari_os.tools.arios cortex lens on

The two MCP tools have different requirements. brain.lens(slug) checks the LENS toggle and looks for a local card or matching indexed content. It doesn't call a vision model. brain.see(image_path, mode='visual') checks LENS and the enabled LLM setting, then requires a working vision backend to caption the image.

If the card is missing or the model is unavailable, inspect the returned message. These distinctions come from the checked MCP implementation, resolving the older guides' conflicting prerequisite descriptions.

Choose the embeddings backend deliberately

python3 -m ari_os.tools.arios cortex embeddings auto|google|ollama|off

These values record a preference in the control-panel configuration. The inspected retrieval and indexing paths instantiate an Ollama client directly; this inspection does not establish that they honour auto, the named provider or off. Check the brain guide before relying on the setting. Local storage alone does not establish processing locality.

Inspect the setting. Enable the capability you need. Check what it returns.
Is every advanced feature off by default?
No. In the checked source, KG, EARS and LENS default off, while wander and predictive work default on. Saved settings and environment overrides can change the result.
Does every LENS call need a vision model?
No. brain.lens reads a local or indexed aesthetic card after the LENS gate. brain.see needs LENS, an enabled LLM setting and a working vision backend.
Does plain EARS need vision?
No. Audio transcription and supported transcript retrieval have their own backend requirements. The command prints its result; it does not automatically save a memory.

Read the brain guide for the core memory path. Add the optional surface only when you can say what job it should do and inspect whether it did it.

[ comments ]

guests welcome. members show first in the list

no comments yet. start the thread_

1 / 9