SeekLink

English · 中文

SeekLink is a local semantic search CLI for Markdown vaults. It indexes a folder
of .md files, searches with hybrid keyword + vector retrieval, and returns
line-anchored results that humans and agents can read with simple shell
commands.

It is built for personal knowledge bases, Obsidian-compatible vaults, bilingual
English/Chinese notes, and local agent workflows. It is also a useful search
layer for Markdown wiki patterns such as Andrej Karpathy's
llm-wiki:
an agent can search existing pages, read precise line windows, then update the
wiki without sending the vault to a hosted service.

Everything runs locally. No API key. No cloud search service. No Obsidian plugin
required.

Install

uv tool install seeklink
# or
pip install seeklink

For Apple Silicon reranking support, install the optional MLX extra:

uv tool install "seeklink[mlx]"
# or
pip install "seeklink[mlx]"

SeekLink requires Python's sqlite3 module to be linked against SQLite
3.45 or newer with FTS5 enabled. seeklink status --vault PATH checks this and
prints a clear error if the runtime SQLite is too old.

Quick Start

# 1. Build the index first.
seeklink index --vault /path/to/vault

# 2. Search it.
seeklink search "machine learning" --vault /path/to/vault

Daily use is simpler if you set a default vault:

export SEEKLINK_VAULT=/path/to/vault
seeklink index
seeklink search "agent memory systems"
seeklink get notes/agent-memory-patterns.md:1 -C 20

seeklink search and single-file seeklink index path/to/file.md auto-use a
resident daemon when SEEKLINK_VAULT is set and --vault is not passed. The
daemon keeps the embedder and optional reranker in memory. Full-vault
seeklink index runs in-process so progress stays on stderr and the final
Done: summary stays on stdout. seeklink status and seeklink get always
stay cold-start: status only reads SQLite metadata, and get reads the file
directly from disk.

Output

Text search output is stable:

  SCORE  PATH[:LINE]  TITLE
           <content preview, one line, up to 120 chars>

• PATH is relative to the vault root.
• LINE is 1-indexed and points to the best matching chunk in the current file.
• Exit code is 0 for success, including no results, and 1 for vault/config
  errors or missing files.
• Scores are useful for sorting within one query. Do not compare scores across
  reranker-enabled and reranker-disabled runs.

Use JSON when an agent needs structured output:

seeklink search "agent memory systems" --vault PATH --json
seeklink status --vault PATH --json

Common Commands

Search

seeklink search "query" --vault PATH [options]

Options:

--top-k N          Number of results. Default: 10.
--json             Emit one machine-readable JSON object.
--tags TAG [TAG]   Filter by tags. AND semantics.
--folder PREFIX    Filter by vault-relative folder prefix.
--rerank-k N|auto  Rerank candidate budget. Default: auto.
--no-rerank        Skip cross-encoder reranking for this query.
--title-weight F   Override title/alias/heading channel weight. Default: 1.5.

Get

Read a precise file window without using the database or daemon:

seeklink get notes/spaced-repetition.md
seeklink get notes/spaced-repetition.md:12
seeklink get notes/spaced-repetition.md:12 -l 40
seeklink get notes/spaced-repetition.md:12 -C 20

-l/--lines prints lines starting at LINE. -C/--context prints lines before
and after LINE, grep-style. Path escapes such as ../.. are rejected.

Status

seeklink status --vault PATH
seeklink status --vault PATH --json

Status reports index counts, model names, index-configuration compatibility,
SQLite WAL status, and freshness warnings. It does not load the embedding or
reranking models.

Index

seeklink index --vault PATH
seeklink index path/to/file.md --vault PATH

Full-vault indexing skips unchanged files by content hash unless the stored
index was built with a different embedder, vector dimension, or chunker
configuration, in which case SeekLink rebuilds the derived index contents.
Single-file indexing updates one Markdown file only when the existing index
configuration is compatible.

Daemon

seeklink daemon --vault PATH

You normally do not run this directly. search and single-file index
auto-spawn and auto-restart the daemon when appropriate. Full-vault index
still runs in-process for progress output. Passing --vault to search or
single-file index forces a one-shot cold-start path because the daemon is
bound to one vault at startup.

How Search Works

SeekLink fuses four channels with Reciprocal Rank Fusion:

The default embedder is jinaai/jina-embeddings-v2-base-zh through
fastembed. CJK full-text search uses a jieba FTS5 tokenizer when the local
Python/SQLite build can safely register it; otherwise SeekLink falls back to
SQLite's built-in trigram tokenizer instead of crashing.

The default vector dimension is 768. Advanced custom-embedder experiments can
set SEEKLINK_EMBEDDING_DIM, but it must match the embedder output and requires
a full seeklink index rebuild.

On Apple Silicon, SeekLink can rerank candidates with
mlx-community/Qwen3-Reranker-0.6B-mxfp8 when installed with seeklink[mlx].
Reranking is local and optional; if MLX is unavailable, SeekLink falls back to
first-stage hybrid RRF ranking. Use --no-rerank for one query or set
SEEKLINK_RERANKER_MODEL="" to disable it globally.

Frontmatter

Markdown frontmatter is optional. When present, SeekLink uses it for tags and
aliases:

---
tags: [ai, memory]
aliases: [LLM memory, agent memory]
---

• tags support filtered search: seeklink search "memory" --tags ai
• aliases are indexed for search and used when resolving wikilinks

Storage

SeekLink writes one SQLite database inside the vault:

/path/to/vault/.seeklink/seeklink.db

The database contains source metadata, chunks, FTS5 tables, sqlite-vec vectors,
and a wikilink graph. Delete .seeklink/ and run seeklink index to rebuild.

Supported

Not For

• Hosted or synced multi-user search.
• Non-Markdown sources without conversion.
• A GUI or Obsidian plugin.
• Sub-millisecond search over millions of notes.
• Cloud embedding or reranking APIs.

Agent Notes

Agents can use SeekLink through ordinary subprocess calls:

seeklink status --vault PATH
seeklink index --vault PATH
seeklink search "query" --vault PATH --json
seeklink get PATH:LINE -C 20 --vault PATH

For hot loops, the daemon exposes a length-prefixed JSON protocol over the Unix
socket at ~/.rhizome/seeklink.sock. Most agents should prefer the CLI JSON
surface unless they specifically need socket-level latency.

See llms.txt for the compact agent contract.

Evaluation

Search-quality tests live in tests/blind/; the method is documented in
docs/blind-test.md. Release claims should be backed by
the bundled fixture queries or by clearly labeled private-vault measurements.

Contributing

git clone https://github.com/simonsysun/seeklink
cd seeklink
uv sync --dev
uv run python -m pytest tests/ -q

Keep runtime dependencies small, keep public docs user-facing, and add a
CHANGELOG.md entry for user-visible changes.

License

MIT