Multi-Index Support for Third-Party Libraries #24
Description
Summary
Investigate whether adding support for indexing third-party libraries (manual, path-based) is worth implementing in ast-rag, or if this should be a separate tool.
---
Proposed Feature
Manual library indexing:
1 ast-rag index-lib --name requests --path /path/to/site-packages/requests
2 ast-rag search "session timeout" --lib requests
Key characteristics:
- User explicitly specifies which libraries to index (no auto-scanning)
- Separate indexes from project code
- Search scoped by --lib flag (project-only by default)
---
Research Questions
┌───────────────────────────┬───────────────────────────────────────────────────┐
│ Question │ Why It Matters │
├───────────────────────────┼───────────────────────────────────────────────────┤
│ Quality improvement │ How much does library context help LLM responses? │
│ Code complexity │ How many LOC changes to support multiple indexes? │
│ Storage overhead │ How much disk space per library index? │
│ Search performance │ Does multi-index slow down queries? │
│ In-scope or separate tool │ Should ast-rag handle this, or a companion tool? │
└───────────────────────────┴───────────────────────────────────────────────────┘
---
Evaluation Criteria
Implement in ast-rag if:
- ✅ < 2000 LOC changes to core
- ✅ < 50 MB per typical library index
- ✅ No measurable search slowdown
- ✅ Clear user value (better code completion / search results)
Separate tool if:
- ❌ Requires major refactoring of IndexManager
- ❌ Indexes > 100 MB for common libraries (requests, django)
- ❌ Duplicates work that ripgrep / sourcegraph already do
- ❌ Better as IDE extension (VSCode, JetBrains)
---
Suggested Approach
Phase 1: POC
1. Pick one library (e.g., requests)
2. Manually index
3. Add --lib flag to search CLI
4. Measure: index size, indexing time, search latency
Phase 2: Evaluate
- Does search quality improve for API-related queries?
- Is the complexity acceptable?
- Go / no-go decision
---
Alternatives to Consider
┌────────────────────────────┬───────────────────────────────────────────┬───────────────────────────────────┐
│ Alternative │ Pros │ Cons │
├────────────────────────────┼───────────────────────────────────────────┼───────────────────────────────────┤
│ Global library cache │ One index per lib, shared across projects │ More complex management │
│ Lazy on-import indexing │ Automatic when code imports lib │ Magic behavior, harder to control │
│ Separate `ast-rag-libs` tool │ Keeps core simple │ Another tool to maintain │
│ IDE integration │ Better UX (click-to-navigate) │ Requires IDE plugin work │
└────────────────────────────┴───────────────────────────────────────────┴───────────────────────────────────┘
---
Deliverables
- [ ] POC branch with single-library support
- [ ] Benchmarks: index size, search latency
- [ ] Writeup: recommended approach (in-core vs. separate)
- [ ] If go: follow-up issue for full implementation
---
Labels