FoundationAgents/MetaGPT · error · Exception

Table not created yet, please add data first.

Error message

Table not created yet, please add data first.

What it means

LanceStore.search raises a plain Exception when self.table is None. The table is only created lazily by LanceStore.add (which calls db.create_table on the first insertion), so searching a store that never received data — or one whose table was dropped — fails with 'Table not created yet, please add data first.'.

Source

Thrown at metagpt/document_store/lancedb_store.py:29

import lancedb


class LanceStore:
    def __init__(self, name):
        db = lancedb.connect("./data/lancedb")
        self.db = db
        self.name = name
        self.table = None

    def search(self, query, n_results=2, metric="L2", nprobes=20, **kwargs):
        # This assumes query is a vector embedding
        # kwargs can be used for optional filtering
        # .select - only searches the specified columns
        # .where - SQL syntax filtering for metadata (e.g. where("price > 100"))
        # .metric - specifies the distance metric to use
        # .nprobes - values will yield better recall (more likely to find vectors if they exist) at the expense of latency.
        if self.table is None:
            raise Exception("Table not created yet, please add data first.")

        results = (
            self.table.search(query)
            .limit(n_results)
            .select(kwargs.get("select"))
            .where(kwargs.get("where"))
            .metric(metric)
            .nprobes(nprobes)
            .to_df()
        )
        return results

    def persist(self):
        raise NotImplementedError

    def write(self, data, metadatas, ids):
        # This function is similar to add(), but it's for more generalized updates
        # "data" is the list of embeddings

View on GitHub (pinned to 11cdf466d0)

Solutions

1. Call store.add(...) (or add_n_docs / your ingestion path) at least once before search so the table is created via db.create_table.
2. If data was previously persisted, reopen the table explicitly instead of relying on lazy creation: store.table = store.db.open_table(store.name) before searching.
3. If operating on a shared db, verify the table exists: if name in [t.name for t in db.table_names()]: store.table = db.open_table(name).
4. Check for empty store at the application level and skip the search when nothing has been ingested yet.

Example fix

# before
store = LanceStore(db, "docs")
results = store.search(query_vec)  # Exception: Table not created yet

# after
store = LanceStore(db, "docs")
if "docs" in db.table_names():
    store.table = db.open_table("docs")
else:
    store.add("", query_vec, {})  # creates table on first add
results = store.search(query_vec)

def ensure_lance_table(store, db):
    """Reopen or create the table so search() has something to query."""
    if store.table is None:
        if store.name in db.table_names():
            store.table = db.open_table(store.name)
        else:
            raise ValueError(f"no data ingested for table '{store.name}'; call add() first")

try:
    results = store.search(qvec, n_results=5)
except Exception as e:
    if "Table not created yet" in str(e):
        raise RuntimeError("vector store is empty; ingest documents before querying") from e
    raise

When it happens

Trigger: Instantiating LanceStore(db, name) and calling search(query_vector, ...) before any successful add() call; or calling search after drop() removed the underlying .lance directory while the instance still exists.

Common situations: Fresh vector store in a new session being queried before ingestion; a crashed/interrupted first add() that never created the table; mixing a persisted db path with a new LanceStore object and assuming the table auto-reopens (it does not — this wrapper keeps table=None until add).

Related errors

• Table not created yet, please add data first
• please check QdrantConnection.
• Delete collection {collection_name} failed.
• Missing fields: {missing_fields}
• use `review` after `fill`

AI-assisted analysis of FoundationAgents/MetaGPT@11cdf466d0 (2026-08-14). Data as JSON: /api/errors/8092f70ea5b14dec. Report an issue: GitHub.