Skip to main content
PAS
Available now

Find, summarize, and manage documents—local search by default

Full-text search and retrieval with local LLM summarization. Writer tools for document creation and indexing require approval gates.

Run a local searchSchemas & examples

What Document Tools do

Document Tools provide full-text search, retrieval, and summarization over your local documents. They work seamlessly with vector memory for RAG-style workflows. Writer tools let you create new documents or index files, but require approval to prevent accidental overwrites.

Reader capabilities

  • Full-text search with filters (date, author, tags)
  • Document retrieval with metadata
  • Chunk summarization via local LLM
  • Integration with vector memory for semantic search

Writer capabilities Gated

  • Write new documents (requires approval)
  • Index files and blobs (gated)
  • Update or delete documents (gated)
Local-first by default: All search and retrieval happens locally. Writes require explicit approval to prevent accidental data loss.

Who benefits from Document Tools

Individuals

Turn inbox clutter and meeting notes into actionable summaries

Example: "Summarize my weekly meeting notes into bullet points" — search, retrieve, and summarize entirely offline.

Teams & Managers

Draft SOPs and policy updates with review workflows

Example: Generate a new SOP from templates and existing docs, then pause for approval before saving.

Developers & IT

Predictable schemas for search and writes; RAG-ready

Example: Integrate document search into internal tools with stable REST endpoints and structured responses.

Security & Compliance

Write gates and explicit ingest allowlists

Control: Require approval for all document writes. Audit logs show who created or indexed what and when.

How it works

1

Search: Full-text or semantic

Use docs.search_docs with keywords or filters. Optionally combine with vector search for semantic recall.

2

Retrieve: Get document content

Call docs.get_doc to fetch full content with metadata (author, date, tags).

3

Summarize: Local LLM processing

Pass retrieved content to the LLM engine for chunk summarization or question answering.

4

Optional: Write or index (gated)

Use docs.write_doc or docs.index_blob to create or index documents. Both require approval.

Safety & approvals: All writes pause for explicit consent. Index paths are allowlisted to prevent accidental ingestion of sensitive files.

Example workflows

Summarize monthly meeting notes

Runs entirely offline
Input:

"Summarize all meeting notes from the last 30 days"

Steps:
  1. docs.search_docs (filter: last 30 days, tag: "meeting")
  2. docs.get_doc (retrieve each note)
  3. llm.generate (summarize key points)
Output:

Bulleted summary with action items—no network calls, fully private

Draft SOP update

Approval before saving
Input:

"Update the onboarding SOP to include new security training"

Steps:
  1. docs.search_docs (find existing onboarding SOP)
  2. docs.get_doc (retrieve current version)
  3. llm.generate (draft updated section)
  4. docs.write_doc (save new version) — pauses for approval
Output:

Updated SOP with tracked changes and approval timestamp

Private RAG over local docs

Runs entirely offline
Input:

"What did we decide about the Q3 roadmap?"

Steps:
  1. llm.embed (query text)
  2. vector.search (find relevant docs)
  3. docs.get_doc (retrieve top matches)
  4. llm.generate (answer with context)
Output:

Answer with citations—all embeddings and generation happen locally

Technical details

Key tools

  • docs.search_docs
  • docs.get_doc
  • docs.write_doc (gated)
  • docs.index_blob (gated)
  • docs.delete_doc (gated)
View tool schemas

Configuration

  • INDEX_PATHS — directories to index
  • MAX_BLOB_SIZE — max file size for indexing
  • CHUNKING_PARAMS — chunk size and overlap
  • WRITE_ENABLE — false (default)

Performance notes

  • Search: 50-200ms for medium document collections
  • Retrieval: 10-50ms per document
  • Indexing: depends on file size and chunking params

Observability

  • Index size and document count
  • Query latency and result counts
  • Write approvals and rejections
  • Indexing throughput

Security posture

Writes disabled by default

All write operations require explicit enablement and approval. Read-only mode is the default.

Explicit ingest allowlists

Indexing is scoped to allowlisted paths. Attempts to index files outside the allowlist are rejected.

Audit logs

Every search, retrieval, write, and index operation is logged with timestamps and approval decisions.

Local-first by default

All search and retrieval happens locally. No network calls unless explicitly configured.

Roadmap & status

Available

Current features

  • Full-text search and retrieval
  • Chunk summarization via local LLM
  • Gated writes and indexing
Planned

Coming soon

  • Advanced ingestion pipelines (PDF, DOCX, etc.)
  • Export formats (Markdown, JSON, CSV)
  • Document versioning and change tracking
View full roadmap

Frequently asked questions

Ready to search and manage documents locally?

Get started with Document Tools in minutes. Everything runs on your machine by default.

Run a local searchView examplesTalk to us