# LanceDB L4 · Runs on its own

LanceDB is a database for AI search.

**Can an AI agent use LanceDB? Yes, without signing up.** Our agent got into LanceDB on its own in 1 minute and did all 8 of its core tasks.

Category: Vector databases · Connects by: MCP, API, SDK · Agent payments: None found · Driven: 30 Oct 2026

**How to connect**
- MCP server URL: `https://docs.lancedb.com/mcp` (streamable HTTP, tested in our drive). Source: https://docs.lancedb.com/mcp
- Docs page: `https://docs.lancedb.com/` (tested in our drive). Source: https://docs.lancedb.com/

**Level 4: Runs on its own.** An agent does the core work and can stop paying on its own.
- Got in by: No key needed (1 minute, no account)
- Got as far as: Use: all 8 core tasks (A person has to buy)
- To reach L5: Pay without a person.

[What the levels mean](https://agentdrives.com/method/levels?via=md)

## From finding it to leaving

What an agent managed without a person, at each step.

1. **Find**: On its own. Docs it can act on: llms.txt, an OpenAPI and an MCP server. (389 doc pages read)
2. **Get in**: On its own. Got in without signing up. (1 minute)
3. **Use**: On its own. Did: create, change and delete tables and rows; search by meaning, by keyword, or both together, with filters. (8 of 8 core tasks)
4. **Pay**: Stopped. Needs a person to buy. Pricing leads to a sales contact form; no key route. Stopped here.
5. **Leave**: On its own. Nothing to cancel, made nothing to delete.

## Getting in

2 of 3 ways in worked without a person.

The way that worked: **No key needed**, 1 minute. No inbox, browser or person needed.

- Agent on a server (API, MCP, CLI): Partly. Without an account (no key needed) it makes its first API call and does all of the core work. Payment: a person has to buy, a sales call.
- Assistant in its own browser (Like ChatGPT agent): Not checked. Sign-up wasn't tried in this drive.
- Signed agent (Known to Cloudflare): Yes. The sign-up page opens.

| Way in | What happened |
|---|---|
| No key needed | Worked: Part of the work done |
| Runs locally, no account | Worked |
| Added by an admin | Not tried: Needs a person's account |

## What an agent can do with LanceDB

All 8 core tasks done, and 15 of 15 others.

- Reliability: core tasks that worked on all 3 runs: 100 of 100
- Handles mistakes: clear errors, retries without duplicates, honest limits: 84 of 100
- Surface tested: documented routes, tools and commands run and checked: 45 of 100
- Docs accuracy: documented operations that behaved as the docs say: 100 of 100
- Docs completeness: how fully the docs describe each operation: 78 of 100

Core tasks:

| Task | Result | Done through | Offered, not tried |
|---|---|---|---|
| Create a table | Done | API | – |
| Insert or upsert records into a table | Done | API | – |
| Delete rows from a table | Done | API | – |
| Run a vector (ANN) search | Done | API | – |
| Run a full-text (BM25) search | Done | API | – |
| Query records with metadata filters / SQL | Done | API | – |
| Create a vector, FTS, or scalar index on a table | Done | API | – |
| Branch a table's history | Done | API | – |
| Run distributed feature-engineering backfills with Geneva | Paid plan: Enterprise-only, sold through sales. | – | API |

15 of 15 other tasks done:

- Combine vector and full-text search with reranking (hybrid search): Done
- Evolve a table's schema (add, alter, drop columns): Done
- List tables and namespaces and read a table's schema: Done
- Restore a table to a previous version or tag: Done
- Create and manage version tags on a table: Done
- Store multimodal blobs (images, audio, video) as table columns: Done
- Create text/image embeddings via the embedding registry: Done
- Rerank search results with a built-in or custom reranker: Done
- Quantize a vector index (PQ, SQ, RaBitQ): Done
- Incrementally reindex a table after writes: Done
- Stream training data batches from a table for PyTorch: Done
- Sync external data sources (Parquet, Lance, IPC, pandas, PyArrow) into a table: Done
- Query a Lance table directly with DuckDB SQL: Done
- Browse ready-to-use Lance-format public datasets: Done
- Monitor object-store request metrics via OpenTelemetry: Done
- Build a vector index on GPU: Paid plan
- Scale Enterprise compute and cache capacity: Paid plan
- Authenticate to LanceDB Enterprise with an API key or OAuth: Paid plan. The sign-up page says sign-ups are currently disabled.

Paid-plan tasks don't count toward the level. The drive used the free plan (LanceDB OSS, no account).

What it can't do, and why:

- Sign up for LanceDB Cloud: The sign-up page says sign-ups are currently disabled.
- Call the hosted REST API: Only Enterprise deployments serve it, set up by the LanceDB team.
- Run Geneva feature backfills or GPU indexing: Enterprise-only, sold through sales.
- Buy a plan or get an API key on its own: Pricing leads to a sales contact form; no key route.

**Surface audit:** 125 documented operations, 56 run, 56 passed, 68 blocked (sandbox or plan limits), 1 left out to stay safe.

## How an agent connects

MCP, API, SDK and llms.txt found.

Agent surfaces: API documented · MCP tested · CLI none found · SDK tested · Docs tested · Agent payments none found. Issues found: 2.

- **MCP server**: 4 tested. Official server. Docs server, 3 tools, open, 28 Jul 2026. https://docs.lancedb.com/mcp (5 tools across its docs) 3 tools listed without signing in on 31 Oct 2026: `search_lance_db`, `query_docs_filesystem_lance_db`, `submit_feedback`.
- **API**: Listed, not tested. Lance REST Namespace, 54 operations; served by Enterprise deployments only. (54 operations across its docs)
- **OpenAPI**: Unknown.
- **CLI**: None found.
- **SDK**: 52 tested. @lancedb/lancedb 0.41.0 (npm), lancedb 0.41.0 (PyPI), Rust crate. (63 methods across its docs)
- **llms.txt**: Yes. https://www.lancedb.com/llms.txt
- **Agent payments**: None found. The checkout wasn't reached: Enterprise.
- **Docs**: No gaps in what we read. Covers getting started, auth, error codes and code examples. 0 of 9 documented ways to do the core work worked when tried. https://docs.lancedb.com/ (389 of 390 pages read)

[The standards](https://agentdrives.com/standard)

## Tips for agent builders

- Use the embedded library: npm install @lancedb/lancedb (or pip install lancedb) and lancedb.connect("./data"); no account is needed
- Do not use the bare relative path "db": connect("db") is treated as LanceDB Cloud and asks for an api_key; use ./db
- Use mergeInsert("id").whenMatchedUpdateAll.whenNotMatchedInsertAll for safe retries; add duplicates rows
- Call table.optimize after writes to fold new rows into vector and FTS indexes; indexStats.numUnindexedRows shows the backlog
- Query the open docs MCP at https://docs.lancedb.com/mcp; its filesystem tool holds the OpenAPI at /openapi/api-reference/rest/openapi.yml (output capped near 30,000 characters per call)

## How LanceDB compares

In Vector databases: 1st of 3 Vector databases products our agent got into on its own.

- **LanceDB** (this page): L4, Gets in on its own
- [Zilliz](https://agentdrives.com/tools/zilliz.md?via=md): L4, Gets in on its own
- [Weaviate](https://agentdrives.com/tools/weaviate.md?via=md): L4, Gets in on its own
- [Chroma](https://agentdrives.com/tools/trychroma.md?via=md): L1, Can't sign up on its own
- [Pinecone](https://agentdrives.com/tools/pinecone.md?via=md): L1, Can't sign up on its own
- [turbopuffer](https://agentdrives.com/tools/turbopuffer.md?via=md): L1, Can't get an API key on its own

[All 6 in Vector databases](https://agentdrives.com/categories/vector-databases?via=md)

Sorted by level, then work done, then time to the first call. [How we rank](https://agentdrives.com/method?via=md#order)

On our standard (25 criteria: against other vector databases products where we measured 5 or more, otherwise against every product we measured):

- [One way to page through lists](https://agentdrives.com/standard#api-03): **Ahead of most** 78 of 87 others (every product) lower, 9 the same.
- [The best way in is built for agents](https://agentdrives.com/standard#in-12): **Ahead of most** 4 of 5 others (vector databases products) lower, 1 the same.
- [Typed, tight input schemas](https://agentdrives.com/standard#mcp-06): **Ahead of most** 3 of 4 others (vector databases products) lower, 1 the same.
- [Descriptions that say when to use the tool](https://agentdrives.com/standard#mcp-04): **In the middle** 2 of 4 others (vector databases products) lower, 2 the same.
- [Agents pick the right tool](https://agentdrives.com/standard#mcp-05): **In the middle** 2 of 4 others (vector databases products) lower, 2 the same.
- [Truthful annotations](https://agentdrives.com/standard#mcp-07): **In the middle** 2 of 4 others (vector databases products) lower, 2 the same.
- [Docs an agent can navigate](https://agentdrives.com/standard#docs-11): **Level with most** 3 of 5 others (vector databases products) the same, 2 lower.
- [Fast to the first call](https://agentdrives.com/standard#in-03): **Level with most** 3 of 5 others (vector databases products) the same, 2 lower.
- [A request id on every answer](https://agentdrives.com/standard#err-07): **Level with most** 3 of 5 others (vector databases products) the same, 2 lower.
- [Versions an agent can pin](https://agentdrives.com/standard#api-07): **Level with most** 59 of 97 others (every product) the same, 38 lower.
- [Answers sized for an agent](https://agentdrives.com/standard#api-04): **Level with most** 3 of 4 others (vector databases products) the same, 1 lower.
- [Every way in works for an agent](https://agentdrives.com/standard#in-11): **Ahead of most** 3 of 5 others (vector databases products) lower, 2 higher.
- [Errors say what to do next](https://agentdrives.com/standard#err-03): **Level with most** 4 of 5 others (vector databases products) the same, 1 lower.
- [A path written for agents](https://agentdrives.com/standard#docs-10): **Level with most** 3 of 5 others (vector databases products) the same, 1 lower, 1 higher.
- [Tool names that sort themselves](https://agentdrives.com/standard#mcp-03): **The same as all** 4 others (vector databases products).
- [A tool list that fits](https://agentdrives.com/standard#mcp-11): **The same as all** 4 others (vector databases products).
- [Real status codes](https://agentdrives.com/standard#err-01): **The same as all** 5 others (vector databases products).
- [Spend an agent can see and cap](https://agentdrives.com/standard#pay-05): **The same as all** 5 others (vector databases products).
- [WebMCP tools on the web app](https://agentdrives.com/standard#proto-02): **The same as all** 5 others (vector databases products).
- [Pages fit one fetch](https://agentdrives.com/standard#docs-08): **Level with most** 4 of 5 others (vector databases products) the same, 1 higher.
- [Results sized for a context window](https://agentdrives.com/standard#mcp-09): **Behind most** 16 of 31 others (every product) higher, 11 the same, 4 lower.
- [Errors a program can read](https://agentdrives.com/standard#err-02): **Behind most** 3 of 5 others (vector databases products) higher, 2 the same.
- [Structured results](https://agentdrives.com/standard#mcp-08): **Behind most** 22 of 36 others (every product) higher, 14 the same.
- [Export before leaving](https://agentdrives.com/standard#leave-03): **Behind most** 4 of 5 others (vector databases products) higher, 1 the same.
- [Prices an agent can read](https://agentdrives.com/standard#find-06): **The lowest result** All 5 others (vector databases products) higher.

[How each criterion is graded](https://agentdrives.com/standard)

## For the team behind LanceDB

**6 things to fix for agents.**

Agent blockers: 2 found · Agent surfaces: no issues found. Safety review: 2 issues found.

Agents get stuck at Pay. Prove LanceDB is yours, then see where each issue is, read your safety findings and ask for a free drive after a fix. How to fix each one comes with the Audit.

- Find: Clear
- Get in: Clear
- Use: Clear
- Pay: Blocks agents
- Leave: Clear

[Claim LanceDB, free](https://agentdrives.com/app/products?add=lancedb.com&ref=listing) · [What a drive covers](https://agentdrives.com/method?via=md)

## Where LanceDB fits

- Category: [Data and storage](https://agentdrives.com/categories/data?via=md), [Vector databases](https://agentdrives.com/categories/vector-databases?via=md)
- Tasks: [Create or update a record](https://agentdrives.com/tasks/create-or-update-a-record), [Query records with filters](https://agentdrives.com/tasks/query-records-with-filters), [Delete a record](https://agentdrives.com/tasks/delete-a-record), [List databases and read their schema](https://agentdrives.com/tasks/list-databases-and-read-their-schema), [Query a database over a REST or GraphQL API](https://agentdrives.com/tasks/query-a-database-over-a-rest-or-graphql-api), [Run a vector search](https://agentdrives.com/tasks/run-a-vector-search)
- Wall: [Let AI agents sign up and get an API key](https://agentdrives.com/guides/let-agents-sign-up-and-get-a-key?via=md) (fix guide)
- Standards: [MCP](https://agentdrives.com/standards/mcp) (tools an agent can call), [llms.txt](https://agentdrives.com/standards/llms-txt) (its docs, listed for agents)
- Steps: [Find](https://agentdrives.com/access/find), [Get in](https://agentdrives.com/access/get-in), [Use](https://agentdrives.com/access/use), [Pay](https://agentdrives.com/access/pay) (stopped here), [Leave](https://agentdrives.com/access/leave)
- Rivals: [Weaviate](https://agentdrives.com/tools/weaviate?via=md) (L4), [Zilliz](https://agentdrives.com/tools/zilliz?via=md) (L4), [Chroma](https://agentdrives.com/tools/trychroma?via=md) (L1)

## Cited in 1 article

- [Can AI agents pay for software? 28 products offer a way](https://agentdrives.com/research/can-ai-agents-pay-for-software?via=md)

Data: [JSON](https://agentdrives.com/api/tools/lancedb.json?via=md) · [Markdown](https://agentdrives.com/tools/lancedb.md?via=md) · [History](https://agentdrives.com/tools/lancedb/history?via=md) · [Atom](https://agentdrives.com/feeds/tools/lancedb.atom) · [JSON Feed](https://agentdrives.com/feeds/tools/lancedb.json) · licence [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)

**Found an error?** [Tell us on this page](https://agentdrives.com/tools/lancedb?via=md#correct) or read [how corrections work](https://agentdrives.com/corrections?via=md).