lancedb

lancedb is a skill for Claude Code, Codex from nebius/nebius-physical-ai. It costs 46 tokens per session (833 once invoked), scanned A, original, Apache-2.0.

A guide for using LanceDB, a database designed for searching records that include data such as images, annotations, and numerical vectors.

In plain words
What is it for?
Use it to create and query tables, import LeRobot or BDD100K datasets, add calculated fields, build materialized views, and work with CLIP image embeddings.
Why use it?
It helps keep dataset imports, derived fields, reusable query views, and searches consistent across command-line tools, software libraries, and services.

Skill for Claude CodeCodex

Written for no agent in particular: nothing here depends on one.

Good fit Use it to create and query tables, import LeRobot or BDD100K datasets, add calculated fields, build materialized views, and work with CLIP image embeddings.

Compare 6 skills from other repositories ↓
Install with agentmods
npx agentmods add skills/nebius/nebius-physical-ai/lancedb
Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

Any agent
npx skills add nebius/nebius-physical-ai --skill lancedb
Clone the repo
git clone --depth 1 https://github.com/nebius/nebius-physical-ai

Made for: Claude Code, Codex.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for lancedb

README.md
[![agentmods](https://agentmods.dev/badge/skills/nebius/nebius-physical-ai/lancedb/github.svg)](https://agentmods.dev/skills/nebius/nebius-physical-ai/lancedb)
Your own site
<a href="https://agentmods.dev/skills/nebius/nebius-physical-ai/lancedb"><img src="https://agentmods.dev/badge/skills/nebius/nebius-physical-ai/lancedb/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for lancedb

Your own site · 80×15
<a href="https://agentmods.dev/skills/nebius/nebius-physical-ai/lancedb"><img src="https://agentmods.dev/badge/skills/nebius/nebius-physical-ai/lancedb.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 46 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 833 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe. Third-party audits
  • NVIDIA SkillSpector pass 7 Sept 2026
How audits are shown
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00046 $0.00833
Opus 5 $0.00023 $0.00417
Sonnet 5 $0.00009 $0.00167
Haiku 4.5 $0.00005 $0.00083

Measured 9d ago against content hash 5ddee0232ce8, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-09, from the pricing page.

Security

Grade A, and why

lancedb scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 9d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

skills/tools/lancedb/SKILL.md · 90 lines

How it starts

The opening of the file, as written. The whole thing — 90 lines — stays where its author put it; the contents beside it link to each section on GitHub.

LanceDB

When To Use

Use this skill for vector-search workbench changes, perception dataset imports, BDD100K failure-mode slices, materialized views, CLIP embedding backfills, and LanceDB CLI/API/SDK parity reviews.

Procedure

  1. Pick the data shape first. LanceDB is best for frame-aligned records such as image paths, annotations, metadata, and vectors. It is not the right store for raw multi-rate sensor streams.

  2. Create or inspect tables before ingestion:

    npa workbench lancedb create-table --help
    npa workbench lancedb query --help
    
  3. Import supported datasets through current commands:

    npa workbench lancedb import-lerobot --help
    npa workbench lancedb import-bdd100k --help
    
  4. Add derived fields through backfill, then materialize reusable SQL slices with create-mv, refresh-mv, and query-table.

Three-Tier Contract

  • CLI: deploy, status, list, create-table, query, import-lerobot, import-bdd100k, backfill, create-mv, refresh-mv, and query-table.
  • SDK/API: keep table import, backfill, and query behavior in shared implementation paths so CLI, SDK, and service endpoints produce equivalent manifests and row counts.
  • YAML: workflow tasks should pass S3-backed LanceDB URIs and table names through environment variables, not hardcoded project paths.

BDD100K Contract

BDD100K UDFs:

  • has_person
  • has_rider
  • person_bbox_area_pct
  • dhash
  • is_duplicate
  • clip_embedding

PERSON_CATEGORIES = {"person", "pedestrian"}. Real BDD100K uses pedestrian; synthetic data may use person. Both must be accepted.

Materialized views are SQL-defined failure-mode slices such as rider_train, nighttime_person_train, and distant_person_train. CLIP embeddings are 512-dimensional float32, use a GPU UDF, and route to H100.

The current GPU-capable image is npa-lancedb:cuda13-b300-0.30.3-sm80-sm90-sm100-sm103-sm120-20260803T031514Z (sha256:a303b53d0769e612101d468ba957656997838f4b8e6a03430f2b5a2c89e3f8b5 in both registries). It was measured on physical B200, B300, H100, and RTX PRO 6000 by embedding three images with CLIP, checking normalized distinct vectors, writing them to Lance, and checking top-1 self-search. The historical 0.30.3 image predates the current transformers return-type fix and must not be used for the CLIP path.

Read the full file on GitHub · 90 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 9d ago First seen · 90 lines · 46 tokens per session scan A 5ddee0232ce8

Subscribe to this mod's changes

lancedb is a skill published in the GitHub repository nebius/nebius-physical-ai (27 stars, last pushed today), licensed Apache-2.0. It adds 46 tokens to every session and 833 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.