Lode

No need to build another knowledge base. Your workspace is already a lode of knowledge.

Lode is a local-first knowledge mining engine that turns your workspace into a searchable knowledge lode.

Why Lode?

Your knowledge already exists.

It lives in:

• Documentation explaining concepts and decisions.
• Markdown notes capturing project knowledge.
• Design documents describing systems and workflows.
• Text files accumulated throughout development.

The problem is not storing more knowledge.

The problem is finding the right piece of knowledge when you need it.

Lode helps AI agents access workspace-specific knowledge by indexing existing documents and exposing them through semantic and lexical retrieval.

Instead of relying only on general-purpose models, agents can access the information that belongs to your projects.

Install

Install from source

1. Clone the repository:

   git clone https://github.com/cubelitblade/lode.git
   cd lode
   

2. Install:

   uv tool install .
   

3. Verify:

   lode --help
   

Quickstart

1. Provide an embedding server

Lode does not ship with an embedding model. Bring your own embedding provider. Lode supports local embedding servers such as Text Embeddings Inference (TEI), as well as OpenAI-compatible embedding APIs.

2. Configure Lode

Set your embedding provider:

lode config set embedding.provider "openai_compatible" --scope user
lode config set embedding.openai_compatible.endpoint <endpoint> --scope user
lode config set embedding.model <model-name> --scope user
lode config set embedding.openai_compatible.key <api-key> --scope user # optional

For a TEI native endpoint, use tei_native instead:

lode config set embedding.provider "tei_native" --scope user
lode config set embedding.tei_native.endpoint <endpoint> --scope user
lode config set embedding.model <model-name> --scope user

Workspace-specific configuration can also be created using workspace scope.

3. Survey a workspace

Discover documents in your workspace:

lode --workspace <path> survey

Survey builds the map of your knowledge lode.

4. Mine a workspace

Generate embeddings and store indexed chunks:

lode --workspace <path> mine

Mining processes discovered documents, generates embeddings, and stores searchable indexes locally in SQLite.

5. Prospect knowledge

Search your workspace:

lode --workspace <path> prospect <query>

Lode performs hybrid retrieval using semantic similarity and BM25 lexical matching.

It combines meaning-based search with exact keyword matching, then returns candidate chunks with scores and digests.

6. Dig the ore

Retrieve a specific chunk:

lode --workspace <path> dig <digest>

This returns the complete content associated with the digest.

FAQ

Why does lode fail to load SQLite extensions on macOS?

Possible reason

Lode requires SQLite extension loading for certain features.

Some Python builds, especially those linked against SQLite libraries without loadable extension support, may not provide sqlite3.Connection.enable_load_extension().

This depends on how Python and SQLite were built, not only on the Python version. CPython documents that loadable SQLite extension support is disabled by default and notes macOS as a notable platform where the underlying SQLite library may lack this capability.

You can verify the capability with:

import sqlite3

print(hasattr(sqlite3.Connection, "enable_load_extension"))

Solution

• Use a separately installed Python distribution instead of the system-provided Python.
• In CI environments, explicitly configure the Python version and interpreter used by uv to avoid relying on the runner's preinstalled Python.

For more information, see:

• https://github.com/python/cpython/blob/main/Doc/library/sqlite3.rst
• https://github.com/python/cpython/blob/main/Doc/using/configure.rst