sia-code 0.7.2

Local-first codebase intelligence with semantic search, multi-hop research, and 12-language AST support

Sia Code

Local-first codebase intelligence for CLI workflows.

Sia Code indexes your repo and lets you:

• search code fast (lexical, semantic, or hybrid)
• trace architecture with multi-hop research
• store/retrieve project decisions and timeline context

Why teams use it

• Works directly on local code (.sia-code/ index per repo/worktree)
• Great for symbol-level search (--regex) and architecture questions (research)
• Supports 12 AST-aware languages (Python, JS/TS, Go, Rust, Java, C/C++, C#, Ruby, PHP)
• Integrates well with LLM CLI agents

Install

# pip
pip install sia-code

# pip with MCP server support
pip install "sia-code[mcp]"

# or uv tool
uv tool install sia-code

# MCP entrypoint from PyPI/local checkout
uv tool install "sia-code[mcp]"

# verify
sia-code --version
sia-code-mcp --help

Quick Start (2 minutes)

# in your project
sia-code init
sia-code index .

# search
sia-code search --regex "auth|login|token"

# architecture trace
sia-code research "how does authentication work?"

# index health
sia-code status

Command Cheatsheet

Command	What it does
sia-code init	Initialize .sia-code/ in current project
sia-code index .	Build index
sia-code index --update	Incremental re-index
sia-code index --clean	Rebuild index from scratch
sia-code search "query"	Hybrid search (default)
sia-code search --regex "pattern"	Lexical search
sia-code research "question"	Multi-hop relationship discovery
sia-code memory sync-git	Import timeline/changelog from git
sia-code memory search "topic"	Search stored project memory
sia-code memory working-set "query"	Build shared working-memory JSON for agents
sia-code memory trace "query"	Trace likely causal timeline events for query
sia-code config show	Print active configuration

Search Modes (important)

• Default command is hybrid: sia-code search "query"
• Lexical mode: sia-code search --regex "pattern"
• Semantic-only mode: sia-code search --semantic-only "query"

Use --no-deps when you want only your project code.

Git Sync Memory + Semantic Changelog

sia-code memory sync-git is the fastest way to build project memory from git history.

• Scans tags into changelog entries
• Scans merge commits into timeline events
• Stores files_changed and diff stats (insertions, deletions, files)
• Optionally enhances sparse summaries using a local summarization model

How semantic summary generation works:

1. sync-git collects git context (tags, merges, commit ranges, diff stats)
2. It gathers commit subjects for each release/merge window
3. A local model (default google/flan-t5-base) generates a concise summary sentence
4. The enhanced summary is stored in memory and later exposed by memory changelog

sia-code memory sync-git
sia-code memory working-set "auth flow" --agent planner --session-id ses-123
sia-code memory trace "why did command parsing change" --format table
sia-code memory add-decision "Keep sqlite-vec default" \
  -d "Need a stable local-first backend baseline" \
  -r "Consistent behavior across environments" \
  --link-file sia_code/config.py \
  --link-symbol default_backend \
  --link-timeline "feature/sqlite-vec->main"
sia-code memory changelog --format markdown

LLM CLI Integration

Primary integration is now the packaged MCP server:

• sia-code-mcp

For engineering workflows, the preferred first MCP call is:

• engineering_bootstrap

It bundles readiness checks, lightweight search, optional memory retrieval, and guarded multi-hop research so MCP-aware clients can use Sia Code effectively with only a bare MCP server configuration.

Integration guide:

• docs/MCP_INTEGRATION.md
• docs/LLM_CLI_INTEGRATION.md

Fallback skill file for environments without MCP support:

• skills/sia-code/SKILL.md

Configuration

Config path:

• .sia-code/config.json

Useful commands:

sia-code config show
sia-code config get search.vector_weight
sia-code config set search.vector_weight 0.0

Note: backend selection is auto by default (sqlite-vec for new indexes, legacy usearch supported).

Multi-Worktree / Multi-Agent Setup

Yes - Sia Code works with multiple git worktrees and multiple LLM CLI instances.

Scope resolution order:

1. SIA_CODE_INDEX_DIR (explicit path override)
2. SIA_CODE_INDEX_SCOPE (shared, worktree, or auto)
3. auto fallback:
   • linked worktree -> shared index at <git-common-dir>/sia-code
   • normal checkout -> local .sia-code

# Shared index across worktrees/agents
export SIA_CODE_INDEX_SCOPE=shared

# Isolated index per worktree
export SIA_CODE_INDEX_SCOPE=worktree

# Full explicit control
export SIA_CODE_INDEX_DIR=/absolute/path/to/sia-index

Typical worktree workflow:

# in main checkout (build shared index once)
export SIA_CODE_INDEX_SCOPE=shared
sia-code init
sia-code index .

# create feature worktree
git worktree add ../feat-auth feat/auth
cd ../feat-auth

# reuse same shared index, then incrementally refresh
sia-code status
sia-code index --update

When a worktree is merged/removed:

• shared scope: index stays in git common dir and remains usable
• worktree scope: index lives in that worktree directory and is removed with it
• after merge, run sia-code index --update in the remaining checkout

Practical guidance:

• Many readers/searchers are fine in shared mode
• Prefer one active index writer per shared index
• For strict branch/agent isolation, use worktree
• Teams on different machines should keep local indexes and sync context via git + sia-code memory sync-git

Documentation

• docs/CLI_FEATURES.md - concise CLI command reference
• docs/MCP_INTEGRATION.md - MCP setup and transport notes
• docs/CODE_STRUCTURE.md - repo/module map
• docs/ARCHITECTURE.md - core runtime architecture
• docs/INDEXING.md - indexing behavior and maintenance
• docs/QUERYING.md - search modes and tuning
• docs/MEMORY_FEATURES.md - memory workflow
• docs/BENCHMARK_RESULTS.md - benchmark summary

License

MIT