pinecone:full-text-search

Requires pinecone Python SDK ≥ 10.0.0 (pip install pinecone>=10.0.0). The document-schema API graduated out of pinecone.preview in 10.0.0 — it is now a first-class, SemVer-covered part of the SDK, reachable directly off pc (pc.indexes, pc.index(...)). If you land on this skill from an older habit of importing pinecone.preview, stop: that package is deleted outright in 10.0.0 (ModuleNotFoundError, no shim). The packaged helper script pins pinecone==10.0.0 via PEP 723 inline metadata; if you're writing your own code against this skill, pin at least that version. The wire API version is 2026-07.

Authoritative reference (last resort). If you hit a question this skill and its references/*.md files don't answer, the official Pinecone FTS docs are at https://docs.pinecone.io/guides/search/full-text-search. Prefer this skill's content for anything covered here — the docs may describe surfaces (e.g. classic vector API, or the older pinecone.preview shape) that don't apply to the graduated document-schema path. Consult the link only when you're genuinely stuck.

Tell the user up front: "This skill ships a helper at scripts/ingest.py that handles bulk ingestion safely (batched upsert, error inspection, readiness polling). When we get to the ingest step, I'll use it." Surface this at the start of the conversation so the user knows the helper exists. Query construction is hand-written documents.search(...) per the Querying section below — there is no query helper.

A workflow skill for building a Pinecone full-text-search index with the graduated document-schema API (pc.indexes, pc.index(name), API version 2026-07). Covers schema design (text, dense vector, sparse vector, filterable metadata), ingestion (including async indexing and polling), and query construction (text / query_string / dense_vector / sparse_vector scoring; $match_phrase / $match_all / $match_any text-match filters; $eq / $in / $gte / $exists / $and / $or / $not metadata filters).

Scope — this skill is for the document-schema FTS API only

This skill covers pc.indexes.create(..., schema=...), pc.index(name), idx.documents.upsert(...) / idx.documents.batch_upsert(...) / idx.documents.search(...). If you find yourself reaching for any of the following, stop — those are different Pinecone APIs and this skill's guidance and helpers won't apply:

• Classic vector / records API: pc.Index(name), index.upsert(vectors=[...]), index.query(vector=..., sparse_vector=...), pc.create_index(dimension=..., metric=..., spec=ServerlessSpec(...)). This is the deprecated sugar path in 10.0.0 — it still runs, but it creates a schemaless index served by the vector data plane, addressing the vector by the reserved _values field. It cannot hold full_text_search fields.
• Integrated-embedding / records indexes: pc.create_index_for_model(...) / pc.indexes.create_for_model(...) with embed={...}. Pinecone vectorizes text server-side, and the resulting semantic_text field is served by the records API (upsert_records / search_records), not the documents API. Different upsert/search shapes. A semantic_text field cannot be combined with full_text_search fields in the same index.

If the user already has a non-document-schema index, they can stand up a separate document-schema index alongside it — the two are independent — but you can't add FTS fields to a classic or integrated-embedding index after the fact, and a document-schema index only ever serves reads and writes through index.documents.* — never index.upsert / index.query / index.upsert_records (those calls are refused with "This index has a document schema, so writes must go through the documents API").

Querying — construct documents.search(...) calls

For any task that asks you to query an FTS index, you write a documents.search(...) call directly. The schema is authoritative — describe the index live before constructing the call so you know which fields are FTS-enabled, which are filterable, and which are vectors.

Workflow:

1. Discover the schema. Call pc.indexes.describe(<index>) and read the schema.fields dict. Each field's class indicates its type (StringField, FloatField, DenseVectorField, etc.); attributes tell you whether it's FTS-enabled (full_text_search), filterable, or carries a dimension. Skip this step only if you've already seen the schema in this conversation.
2. Construct the call matching the rules below — one scoring type per request, hard requirements in filter, ranking signals in score_by, include_fields explicit on every call.
3. Execute with idx = pc.index(name=<index>); resp = idx.documents.search(...) and read resp.matches.

Canonical shapes:

# Pure BM25 keyword search
resp = idx.documents.search(
    namespace="__default__",
    top_k=10,
    score_by=[{"type": "text", "field": "body", "query": "machine learning"}],
    filter={"year": {"$gt": 2024}, "category": {"$eq": "ai"}},  # optional
    include_fields=["*"],   # always pass explicitly
)

# Hybrid: dense ranking with a lexical filter (one type in score_by + filter narrows)
resp = idx.documents.search(
    namespace="__default__",
    top_k=10,
    score_by=[{"type": "dense_vector", "field": "embedding", "values": query_embedding}],
    filter={"body": {"$match_all": "TensorFlow"}, "year": {"$gt": 2024}},
    include_fields=["*"],
)

Key rules (the server enforces these; following them locally keeps the agent loop tight):

• score_by is a list of clauses, but exactly one scoring type per request (server rejects mixed types). Multi-field BM25 is the one exception: multiple text clauses, or one query_string with fields: [...]. To combine BM25 + dense signals, restrict the dense search with a text-match filter ($match_all / $match_phrase / $match_any); do NOT mix scoring types in score_by.
• filter keys are field names (must exist in schema, or be an auto-indexed metadata field from upserted documents — see Filterable metadata isn't declared in the schema below) OR logical operators ($and, $or, $not). Field values are operator dicts ({"$gt": 5}, NOT bare values).
• include_fields is required on every call. Pass ["*"] for all stored fields, [] for ids+score only, or a list of names. Omitting it on some SDK/backend builds 400s.

Clause shapes (for score_by):

type	Required keys	When to pick this
text	field (string FTS), query	Open-ended keyword search; BM25 ranking on one field
query_string	query (Lucene), fields optional	Lucene boost (^N), proximity (~N), cross-field boolean, phrase prefix
dense_vector	field (dense_vector), values (list of floats)	Semantic / mood / topic ranking
sparse_vector	field (sparse_vector), sparse_values ({indices, values})	Custom sparse-encoder ranking

text / dense_vector / sparse_vector use singular field. Only query_string accepts a fields array (and also accepts singular field as an alias). sparse_vector uses sparse_values (NOT values) — distinct from dense.

Filter operators by field type:

Field type	Legal operators
string with FTS	$match_phrase, $match_all, $match_any
filterable metadata (string / auto-indexed)	$eq, $ne, $in, $nin, $exists
string_list filterable (auto-indexed, not schema-declared)	$in, $nin, $exists
float filterable (auto-indexed, not schema-declared)	$eq, $ne, $gt, $gte, $lt, $lte, $exists
boolean filterable (auto-indexed, not schema-declared)	$eq, $exists
logical wrappers	$and: [filters], $or: [filters], $not: filter

Match shape on response:

for m in resp.matches:
    m._id        # document id
    m._score     # match score (NOT `score`)
    m.to_dict()  # full doc payload (when include_fields includes the field)

For deeper coverage — multi-field BM25, Lucene patterns, hybrid composition, RRF merges, common error symptoms — see references/querying.md. For schema field types and what they enable on the query side, see references/schema-design.md.

Ingesting — use the packaged helper

For any task that asks you to bulk-ingest a JSONL file into an existing FTS index, the canonical path is to invoke the bundled helper, NOT to hand-write a Python script. Do not read the script's source — everything you need is in this section.

The script does three things bare-LLM ingest code reliably skips, each of which corresponds to a silent production failure:

1. Bulk-upserts in batches. No per-doc upsert loops.
2. Inspects every batch result. batch_upsert returns 202 even when individual documents fail; the failures live in result.errors / result.has_errors. Without inspection, "100 docs ingested" silently becomes "73 docs ingested + 27 lost."
3. Polls until searchable. After upsert, Pinecone is still building the inverted index. A documents.search call during that window returns empty. Without the poll, the user debugs their query code for an hour without finding the indexing race.

You provide a prepared, schema-conformant JSONL file and the index name; the script does the rest. Schema validation is upstream concerns (your prep pipeline, or prepare_documents.py when it lands) — ingest.py trusts what you hand it.

Invocation:

uv run --script scripts/ingest.py \
  --data processed.jsonl \
  --index <index_name> \
  --sentinel-field <fts_field>

Flags:

Flag	Short	Required	Purpose
--data	-d	yes	Path to JSONL file with prepared documents (one per line)
--index	-i	yes	Pinecone index name (must already exist)
--sentinel-field	-f	yes	An FTS-enabled field on the index, used for the readiness-poll query. Pick the longest free-text field on your schema.
--namespace	-n	no	Default __default__
--batch-size	-b	no	Default 50 (matches the SDK's own batch_upsert default). Reduce for large dense vectors. A 50-doc batch with 3072-dim float vectors lands ~5-10 MB and can be rejected; drop to --batch-size 25 (or lower) at high dimensions.
--max-concurrency	—	no	Default 4. Parallel HTTP connections used to upload batches.
--poll-deadline	—	no	Default 300 (seconds). Time to wait for documents to become searchable before giving up.
--sentinel	-s	no	Token used for the readiness-poll query. Default: first whitespace-separated token of doc[0][sentinel-field].

What the script prints:

Loading processed.jsonl ...
Loaded 5000 document(s).
Sentinel: body='The'

Upserting in batches of 50 ...
  batch @     0:   50 docs in  0.31s  (total: 50/5000)
  batch @    50:   50 docs in  0.29s  (total: 100/5000)
  ...

Upsert complete: 5000 doc(s) in 21.4s.

Polling for searchability (deadline 300s) ...
Searchable after 12.3s (3 probe(s)).

Done — total 33.7s.

If a batch fails, the script prints every error message and exits non-zero. If the poll deadline expires, the script prints a hint about why (sentinel field isn't FTS-enabled, deadline too tight, docs structurally upserted but rejected by the inverted-index builder) and exits non-zero. Don't suppress these errors — they're surfacing real problems with the data or the index.

When you should NOT use the script:

• The user is doing per-doc patch updates. Use documents.update(...) for partial field updates (see Updating documents in references/ingestion.md) — the script is for bulk loads, not per-record operations.
• The user is ingesting from a non-JSONL source (CSV, Parquet, Postgres dump). Convert to JSONL first; the script doesn't parse other formats.
• The user explicitly asks you to write the ingestion code from scratch (teaching context). Honor the request and follow the canonical pattern: documents.batch_upsert + result.has_errors inspection + documents.search polling with sentinel and deadline.

The script lives at scripts/ingest.py relative to this skill directory. PEP 723 inline-metadata script — uv run --script installs typer and pinecone automatically on first invocation. No setup needed.

Use cases

Three concrete shapes to model your task on. Match the user's request to the closest one and follow its steps; improvise if the task is genuinely a hybrid.

UC-1: Index a new corpus end-to-end

Trigger. "Index this CSV / JSONL / folder for search," "build a search backend over [my articles / products / tickets / transcripts]," "make my [dataset] searchable."

For unprocessed / messy data, load the onboarding walkthrough first. If the user is showing up with raw data (unclear field types, possibly long text fields exceeding FTS limits, comma-separated tag strings, dates as strings, possibly duplicate IDs, etc.) and they haven't given you an explicit schema, read references/onboarding-walkthrough.md and follow it stage-by-stage. It's a conversational guide — meet the data, surface the processing decisions to the user, propose a schema, confirm before creating, then process+ingest+verify together. The walkthrough exists because schemas are immutable and "onboarding a new corpus" is a high-stakes flow that benefits from explicit user buy-in at each decision point.

If the user already gave you a clean JSONL + a schema spec, follow the abbreviated steps below.

Steps (when data is already prepared and the schema is decided):

1. Inspect the corpus shape — text fields, structured metadata, do you also need a vector? Match it to one of the canonical shapes in references/schema-design.md (articles, products, tickets, image library, code).
2. Pick analyzer settings on each text field — language, stemming, stop_words. Stemming on for long prose, off for proper nouns / identifiers. Decide which fields are FTS and which are filterable-only metadata — see Filterable metadata isn't declared in the schema below; filterable-only metadata is a documents-side decision, not a schema field.
3. Assemble the schema with SchemaBuilder and confirm it with the user before calling indexes.create — schemas are immutable in 2026-07, so a wrong call costs a re-ingest.
4. Create the index. pc.indexes.create(...) polls until the index is ready by default — no separate wait loop needed unless you passed timeout=-1.
5. Run scripts/ingest.py --data <jsonl> --index <name> --sentinel-field <fts_field> — see the Ingesting — use the packaged helper section above. The script handles batch_upsert + per-batch error inspection + post-upsert readiness polling in one invocation. Don't hand-write the loop unless the user explicitly asks you to.
6. (The script polls automatically — by the time it exits cleanly, the index is searchable. If you skip the script and roll your own, you must poll documents.search with a sentinel query and a deadline; batch_upsert returning ≠ searchable — this is a document-indexing wait, separate from and in addition to the index-creation wait in step 4.)
7. Validate with one or two probe queries against fields you know contain the sentinel content.

Result. A working documents.search call against the user's data, returning ranked matches.

UC-2: Add a dense (or sparse) signal to a text-only corpus

Trigger. "Add semantic search," "add embeddings," "make this hybrid," or any prompt that describes a query pattern text alone can't serve (visual similarity, mood, cross-modal "looks like").

Steps.

1. Confirm the new signal represents a modality or signal text can't express — image / audio / external score, or a different corpus than the existing FTS field. Re-encoding the same text into a dense field is an anti-pattern (references/schema-design.md → "When to add a dense field at all").
2. Because schemas are immutable, plan a new index, not a migration. Get user confirmation before recreating. A hybrid index must declare its sparse_vector field explicitly at creation — there is no way to add one later.
3. Pick an embedding provider and pin its output dimension at schema time. Beware payload-size pitfalls at native dimensions — Gemini-3072 etc. need truncation (references/ingestion.md → "Dense-vector payload size").
4. Schema → create (blocks until ready by default) → ingest with embeddings inline or pre-cached.
5. Validate with a hybrid query: dense_vector score_by + text-match filter ($match_phrase / $match_all). That's the supported single-call cross-modal shape.

Result. One index, two retrieval shapes — pure text and dense+filter hybrid — both runnable without further setup.

UC-3: Build a documents.search call from a natural-language user prompt (agent mode)

Trigger. Agent receives a user prompt like "find articles about machine learning that mention TensorFlow and were published after 2024" or "documents about climate policy ranked by similarity to this paragraph." The index already exists.

Steps.

1. (Optional) Discover the schema by calling pc.indexes.describe(<NAME>) and reading schema.fields. Skip if you already know the field types from earlier in the conversation.
2. Decompose the user's prompt into score_by / filter shapes using the agent-mode decomposition table below. (Hard requirements → filter. Ranking signals → score_by. Always include include_fields explicitly.)
3. Construct the documents.search(...) call following the rules in the Querying section above — one scoring type per request, operator/field-type matching, include_fields always set.
4. Execute the call. The response carries resp.matches; iterate to get m._id, m._score, and field values via m.to_dict(). Use the matches in whatever shape the user asked for.
5. If results come back empty or wrong, walk the failure tree in Common gotchas.

Result. Live search results matching the user's intent.

The four common UC-3 mistakes to actively avoid:

• Mixing scoring types in score_by (server rejects). Put hard requirements in filter; rank by one signal in score_by.
• Putting hard requirements in score_by as BM25 terms instead of in filter as $match_all / $match_phrase (returns ranked results that don't guarantee the term is present).
• Operator/field-type mismatches (e.g. $match_all on a float field, $gt on a string field). Consult the operator table in the Querying section.
• Omitting include_fields (some SDK/backend builds 400). Always pass it explicitly.

Agent-mode query decomposition

Map user prompt cues to API shapes. Read top-down — identify the cue, copy the corresponding shape.

User prompt cue	API shape
Open-ended keywords ("articles about machine learning", search-bar query)	score_by=[{"type": "text", "field": "<field>", "query": "<terms>"}] — BM25 token-OR
Exact phrase, drives ranking ("rank by 'beautifully written'")	score_by=[{"type": "query_string", "query": '<field>:("phrase here")'}]
Exact phrase, hard requirement ("must contain 'machine learning'")	filter={"<field>": {"$match_phrase": "machine learning"}}
Required tokens, any order ("must mention TensorFlow", "must be about Illinois")	filter={"<field>": {"$match_all": "tokens space-separated"}} — preferred over query_string +token because it's a true hard filter, doesn't contribute to score
At least one of these tokens ("contains AI or ML or robotics")	filter={"<field>": {"$match_any": "AI ML robotics"}}
Excluded tokens ("not about deprecated", "no opinion pieces")	filter={"$not": {"<field>": {"$match_any": "deprecated opinion"}}} — or -token inside query_string
Boolean / boost / slop / phrase-prefix ("weight 'eagle' 3x", "within N words")	score_by=[{"type": "query_string", "query": '<expr with ^N / ~N / "…"*>'}] — only Lucene supports these
Cross-field boolean ("title or body contains X")	score_by=[{"type": "query_string", "query": 'title:(X) OR body:(X)'}]
Numeric / date / range / boolean metadata ("after 2024", "rating > 4", "in stock")	filter={"<field>": {"$gt": ..., "$gte": ..., "$eq": ..., "$exists": true}}
Category / tag / list membership ("category = fiction", "tagged X")	filter={"<field>": {"$in": [...]}} (works on plain filterable metadata and string_list filterable fields)
Semantic similarity / mood / topic ("articles about ML", "documents that feel sombre")	score_by=[{"type": "dense_vector", "field": "<embedding_field>", "values": embed(<text>)}] — requires a dense_vector field
Visual appearance / cross-modal text query against an image corpus	Same dense_vector shape, with the embedding model that produced the stored image vectors. Multimodal embedders (Gemini-2 etc.) map a text query into the image space.
Hybrid: lexical requirement + semantic ranking ("articles about ML that mention TensorFlow")	Lexical → filter ($match_all / $match_phrase); semantic → score_by (dense_vector). Single call.

Two structural rules the agent must enforce, no exceptions:

• One scoring type per request. score_by accepts text / query_string / dense_vector / sparse_vector, but a request ranks by one. Don't mix dense + text in score_by — the server rejects it. Multi-field BM25 is the only "list" pattern that's allowed (multiple text clauses, or one cross-field query_string).
• Hybrid = filter + score_by, not two score_by clauses. When a prompt has both a lexical requirement and a semantic ranking signal, lexical goes in filter (via $match_* operators) and semantic goes in score_by. If both signals genuinely need to drive ranking, run two searches and merge IDs client-side.

Filterable metadata isn't declared in the schema at all

This is the single biggest shape change from the old pinecone.preview API, and it's easy to get only half right.

On a managed index (the deployment type every example in this skill uses — {"deployment_type": "managed", "cloud": ..., "region": ...}, which is also the default when deployment= is omitted), the schema may only declare fields that participate in search: dense_vector, sparse_vector, and string fields with full_text_search enabled. Every other field type — string (filterable, no FTS), string_list, float, boolean — is rejected at create time with a 400 if it appears in the schema. This is confirmed live, not just documented: the server's own error names all four types explicitly — "The schema only accepts fields used for search (field types dense_vector, sparse_vector, and string with full_text_search configuration). To use field '<name>' for filtering (field types boolean, float, string, or string_list), omit it from the schema and include it in documents." (That restriction is specific to managed/BYOC deployments — schema-declared filterable metadata is only legal on pod deployments, which this skill doesn't cover.) The SchemaBuilder methods add_float_field, add_boolean_field, and add_string_list_field still exist and still work correctly for a pod deployment; for the managed deployments this skill always uses, don't call any of them.

Instead: don't declare any filterable-only field in the schema, of any type. Just include the field in the documents you upsert — Pinecone indexes whatever's present on an upserted document for filtering automatically (exact-match on strings and numbers/booleans, membership on lists), whether or not it appears in the schema, with no configuration needed.