Sia Code

Local-first codebase intelligence for CLI workflows.

Sia Code helps you:

• search and research code with lexical, semantic, or hybrid retrieval
• run multi-hop architecture research across files, symbols, and repos
• trace historical file and behavior changes (memory trace, memory git-context)
• inspect blast radius before refactors
• work across big repos, multi-repo workspaces, and git worktrees

Why use it

• Local index. No hosted service required.
• Works well on big repos (think ~5,000 files, not only small projects).
• Aggregates search, research, and status across multiple repos in one workspace.
• Handles both architecture questions and history-aware change questions.
• Good fit for MCP clients and skill-based CLI agents.

Install

# MCP + CLI
uv tool install "sia-code[mcp]"

# CLI only
uv tool install sia-code

If you prefer pip, pip install sia-code and pip install "sia-code[mcp]" also work.

Quick start

1) Skill first

Source skill file:

• skills/sia-code/SKILL.md

Ask your agent to install from the file link.

2) MCP

uv tool install "sia-code[mcp]"
sia-code-mcp

Simple JSON reference:

{
  "mcpServers": {
    "sia-code": {
      "command": "uvx",
      "args": ["--from", "sia-code[mcp]", "sia-code-mcp"]
    }
  }
}

Then point your MCP client at sia-code-mcp. See docs/MCP_INTEGRATION.md and docs/LLM_CLI_INTEGRATION.md.

3) Direct CLI

# initialize + index
uvx sia-code init
uvx sia-code index .

# lexical search for exact symbols
uvx sia-code search --regex "AuthService|token"

# hybrid / semantic-style research
uvx sia-code research "how does authentication work?"

# historical change research
uvx sia-code memory trace "why did auth behavior change" --format table
uvx sia-code memory git-context src/auth.py

Big repos, multiple repos, worktrees

• Big repo: built for real codebases with thousands of files, including ~5,000-file projects
• Multi-repo workspace: run index . from parent folder; Sia Code auto-detects git sub-repos, indexes each, then aggregates search, research, and status
• Git worktrees / multi-agent sessions: use shared or isolated indexes with SIA_CODE_INDEX_SCOPE=shared|worktree

# shared index across worktrees/agents
export SIA_CODE_INDEX_SCOPE=shared

# isolated index per worktree
export SIA_CODE_INDEX_SCOPE=worktree

History-aware understanding

Use Sia Code for both code lookup and change understanding:

• search -> lexical, semantic, or hybrid retrieval
• research -> multi-hop architecture tracing across related files and symbols
• memory sync-git -> imports merges/tags as searchable timeline + changelog context
• memory trace -> explains likely causal history for behavior or symbol changes
• memory git-context <file> -> shows effective file history, likely owners, reverts, evolution narrative, and co-change blast radius
• memory working-set "query" -> emits compact shared JSON context for agent handoff

More docs

• docs/CLI_FEATURES.md
• docs/INDEXING.md
• docs/QUERYING.md
• docs/MEMORY_FEATURES.md
• docs/MCP_INTEGRATION.md
• docs/LLM_CLI_INTEGRATION.md
• docs/ARCHITECTURE.md

License

MIT