Semantic code search plugin for OpenCode.

Add to your opencode.json:

{
  "plugin": ["@jrc2139/opencode-plugin-sema"]
}

The plugin is automatically installed from npm when OpenCode starts.

The sema binary must be installed (version 0.2.0 or later):

curl -fsSL https://sema.sh/install.sh | sh

That's it! Sema automatically builds the index on first search. No manual indexing required.

The plugin provides multiple tools that can be invoked by the LLM:

Use sema to find where authentication is handled

Or run sema directly in the terminal:

# Hybrid search (default) - best for natural language questions
sema "where is authentication handled?"
sema "error handling patterns" -l python

# Keyword search (-k) - fast, no model loading, good for identifiers
sema -k "parseArgs"
sema -k "ConfigLoader"

# Search within a directory
sema "API endpoints" src/
sema "database queries" ./backend/

# With filters
sema "error handling" -l zig -n 10
sema -g "src/**/*.ts" "authentication"
sema --exclude "**/tests/*" "main function"

These commands require a previously built index (built automatically on first search):

Use sema_find to locate the HttpClient class definition
Use sema_refs to show all calls to parseConfig
Use sema_query to find all exported functions with complexity > 10
Use sema_outline to show the structure of src/engine/engine.zig
Use sema_blast to see what would break if I change parseConfig

Terminal usage:

# Find symbol definitions
sema find parseConfig
sema find HttpClient --kind class
sema find parse --prefix  # finds parseConfig, parseArgs, etc.
sema find doWork --exported  # only exported symbols

# Find symbol references
sema refs parseConfig
sema refs HttpClient

# File outline with symbol nesting
sema outline src/engine/engine.zig
sema outline src/engine/engine.zig -v  # include signatures

# Impact analysis
sema blast parseConfig
sema blast handleConnection

# Structural queries
sema query --kind function --min-complexity 10
sema query --kind class --language python
sema query --parent HttpClient  # methods of HttpClient
sema query --role declaration --exported

1. Auto-Index: First search automatically indexes the codebase (no manual sema index needed)
2. Keyword Search: Fast, instant FTS-only index with no model loading (~5ms)
3. Hybrid Search: Shows keyword results immediately while building semantic index in background
4. Server: Runs in background for ~50ms response time, 30-minute idle timeout
5. Plugin Load: Starts sema serve if needed, reuses existing server

By default, sema respects .gitignore files. You can also use .semaignore files for sema-specific exclusions (same syntax as .gitignore):

# .semaignore - exclude from search without affecting git
generated/
*.min.js
vendor/
build/

To ignore gitignore files and index everything:

sema index --no-gitignore .

Or in .sema/config.json:

{
    "respect_ignore": false,
    "db": {
        "mode": "hybrid",
        "chunker": "ast"
    }
}

Optional settings in .sema/config.json:

• db.mode: "hybrid" (default, semantic + keyword) or "keyword" (FTS only, fast)
• db.chunker: "ast" (default, AST-based code splitting) or "raw" (line-based)
• respect_ignore: true (default, respects .gitignore) or false (index everything)

Apache-2.0