Enable the language layer

Optional, and the console is useful without it. Findings, the query workspace and compare mode all work with nothing configured and cost nothing to run.

Two ways in

Locally, free, offline, no account:

bash

ollama pull qwen3:8b

That is the whole setup. The console probes localhost:11434 and comes up local.

Or with Claude:

bash

export ANTHROPIC_API_KEY=sk-…

A key beats a local model when both are present. An explicit choice in Settings → Intelligence beats both.

You can also point at any OpenAI-compatible endpoint — vLLM, LM Studio, Groq, llama.cpp — with a base URL and a model name.

Check it actually works

Settings → Intelligence → Test the model spends one real call and reports the round trip:

answered, and honoured the schema
gemma3:27b · 11.8s · 72 in / 41 out · no cost — this ran on your machine

A failure answers the same way, with the reason: a stale key, a model that was deleted, an endpoint answering prose where a schema was promised. Better a sentence here than a mystery three features later.

What it will do

Translate a question into a filter. Type English above the Rows filter. The predicate lands in the box for you to read — nothing runs until you say so — and it is dry-run counted first, so you see whether it understood you before spending a page read:

"moments in the Go devroom more than ten minutes in"
  → matches 44 of 1,114 rows
  filter box:  track = 'Go' AND ts_s > 600

Refusing is a first-class outcome. Ask for something the columns cannot express and it says so rather than producing a filter that runs and means something else.

Describe a table. Insights → Describe this table writes a few sentences from the schema, the statistics and the console's own findings. The answer is kept against the table version, so it is asked once: 47 seconds the first time, 0.06 the second.

What it costs

Every response carries its own cost, and the Insights tab totals what this server has spent since it started. A cache hit is counted as a call that did not happen. A local model costs zero; a model with no published price reports its cost as unknown rather than as zero, because those are different claims.

Set a ceiling in settings or with LANCESCOPE_SPEND_CEILING. It is checked before each call — refusing after the money is gone is a receipt, not a limit.

What leaves your machine

Schema and statistics only, by default. A table summary is written from the shape of a table, never its contents.

The one exception is deliberate and visible. Translating a question into a filter is much more accurate when the model knows the actual values of low-cardinality string columns — without them it writes track = 'Go devroom' against a corpus whose track is Go. So:

with a local model, those values are sent, because nothing leaves the machine
with a hosted model, they are not, unless you ask for them
either way the response names the columns whose values were sent

Blob columns, vectors and binary columns are never sent under any setting.

Nothing is asked on load

Every model call is something you pressed. A panel that spends money because somebody opened a tab is a panel that spends money nobody agreed to spend.