Local context built for AI agents

Make every file.
Searchable by AI.

Context Mole extracts the text and structure from your documents, spreadsheets, slides, emails, images, and archives—including supported files nested inside them—so your agent can search the right passages instead of repeatedly reading entire files.

Download for Windows View source

FOSS Free, local, read-only, and open source.

Less file loading · More useful context · Better agent retrieval

50+supported file types

Nestedattachments & archives

LocalOCR, embeddings & search

Read-onlyaccess for AI agents

Why agents work better with it

More knowledge.
Less wasted context.

Context Mole prepares your files for agent use before the question is asked. The result is faster discovery, tighter evidence, and less need to push large binaries through an AI conversation just to find the useful part.

01Extract once

Turn files into searchable text

Context Mole extracts embedded text, tables, document structure, and OCR from scans and images, then keeps it in one local index your agent can search again and again.

02Go deeper

Find what is buried inside

Supported content is discovered inside email attachments, Office packages, PDFs, and archives. Each nested item is indexed separately without losing its path back to the source.

03Use less context

Return passages before whole files

Your agent gets focused previews and neighboring text first. It can open an original file or materialize one attachment only when the task actually needs it.

04Search like an agent

Control recall and precision

Keyword, semantic, and hybrid retrieval support exact clauses, exclusions, field weights, filters, and follow-up searches scoped to stable content IDs.

05Keep the proof

Trace every answer to its source

Results preserve the exact file, attachment chain, page, sheet, slide, heading, or message location so agents can inspect and cite the right evidence.

06Share the index

Connect every agent once

All supported AI clients search the same local database and embeddings. You do not need to rebuild or duplicate your knowledge base for every assistant.

Search below the surface

Nested files become
first-class context.

A useful answer may be inside a PDF attached to an email stored in an archive. Context Mole follows that supported content chain, extracts every searchable layer, and preserves the route back to the original source.

01Attachments and archive entries receive their own stable content IDs.
02OCR makes scanned pages and image text searchable alongside native text.
03One broken, encrypted, or unsupported item does not stop the rest of the project.

INDEXED CONTENT CHAIN4 LAYERS / READY

ZIP
client-records.zipRoot document
indexed
MSG
renewal-thread.msgNested email
extracted
PDF
signed-agreement.pdfEmail attachment
OCR + text
P14
Renewal & terminationMatched passage · page 14
returned

client-records.zip › renewal-thread.msg › signed-agreement.pdf › page 14

Make AI usage go further

Stop paying your agent
to re-read everything.

Large PDFs, scans, and folders can consume valuable context before the real reasoning even starts. Context Mole lets the agent retrieve compact passages first and request the original only when necessary—helping reduce repeated file reads and conserve AI usage.

Exact savings depend on the AI client, model, files, and task.

Without an indexHEAVY

Load files. Search manually. Repeat.

Send or open whole PDFs and image files
Spend context finding the relevant page
Repeat the work in the next agent or chat

With Context MoleFOCUSED

Search passages. Inspect the proof. Continue.

Search extracted text across every indexed source
Return a few relevant passages with provenance
Open only the selected file or nested attachment

Three-step workflow

Index once.
Ask naturally.

Context Mole handles the preparation in the background. You keep working in the AI agent you already use.

01
Add folders to a project
Choose the local folders the agent is allowed to search. Context Mole watches them for changes.
02
Extract and index every supported layer
Native extraction, OCR, keyword indexing, and optional embeddings run locally in the background.
03
Let the agent retrieve what it needs
Connect through read-only MCP and search the same index from any supported AI client.

Agent-directed retrieval

Search the way
the task demands.

This is not a generic chat-with-your-files layer. The connected agent controls the retrieval strategy, exact clause logic, ranking, filters, and follow-up scope.

01 / Choose

The agent selects the retrieval mode

Keyword for exact evidence, semantic for concepts, or hybrid when both signals help.

02 / Shape

Clauses and weights stay explicit

Mix required, optional, and excluded terms, phrases, and prefixes. Tune fields and filters for the task.

03 / Follow

Every result can narrow the next step

Focus on stable content IDs, read neighboring passages, or materialize only the selected original content.

search_projectSTRUCTURED / READY

MODE: hybrid
CLAUSES: must phrase · should prefix · must_not term
SCOPE: PDF attachments inside MSG email files
RESULTS: Grouped passages · source chain preserved

FOLLOW-UP

Focus on a returned content_id, then read adjacent passages or materialize only that content.

Broad coverage, one index

All the context
your agent can reach.

Files remain on your computer. Documents are opened read-only, and extraction, OCR, embeddings, and search run locally after model downloads.

01 File types50+ document, data, image, mail & archive types

Supported content is also extracted when it appears inside email attachments, Office packages, PDFs, or archives. Macros are never executed.

Documents

PDF · DOCX / DOCM / DOTX / DOTM · XLSX / XLSM / XLTX / XLTM · PPTX / PPTM / PPSX / PPSM / POTX / POTM · ODT / ODS / ODP · RTF · EPUB

Tables & data

CSV · TSV · JSON · JSONL · XML · YAML / YML · TOML

Text & web

TXT · LOG · Markdown · RST · AsciiDoc · TeX · HTML / HTM · MHT / MHTML

Images

PNG · JPEG · BMP · GIF · WebP · TIFF

Email

EML · MSG

Automatic setup

OpenAI Codex / ChatGPT
Claude Code
Claude Desktop
Cursor
Zed
GitHub Copilot CLI
Gemini CLI
Google Antigravity
Kiro
JetBrains Junie
Devin CLI / Desktop
Windsurf Cascade (legacy)
Cline

Guided setup

Visual Studio Code
Roo Code
OpenCode

05 MCP tools8 read-only tools for discovery and evidence

Connected agents use these tools to discover projects, search the index, inspect evidence, and access authorized original content only when needed.

list_projects
Lists initialized projects, authorized folders, search generation, and document status counts.
search_project
Searches one project with agent-selected keyword, semantic, or hybrid retrieval, structured clauses, weights, filters, and grouped results.
read_passages
Reads the full indexed text for selected passage IDs, with optional neighboring passages for context.
get_document_info
Returns metadata, revision details, extraction counts, and recorded errors for one document or content node.
list_documents
Browses and filters a project’s root-document inventory and indexing status without loading extracted text.
list_attachments
Lists attachment and archive-entry content IDs and their hierarchy for one indexed root document.
resolve_local_file
Resolves an indexed document or content ID to its authorized original source file when that file is sufficient.
materialize_content
Provides an original root file or safely extracts one selected attachment or archive entry for deeper inspection.

Better context for better agents

Make your files searchable.
Keep your context focused.

One private, local index for the AI tools you already use.

Get Context Mole

Make every file.Searchable by AI.

More knowledge.Less wasted context.