v0.1.3
Pre-release
Pre-release
First tagged release. seedloom introspects a live Postgres schema and uses an LLM to generate realistic, referentially-valid seed data - foreign keys are structurally constrained to real inserted parent values, never invented.
Highlights
- pgvector support - detects
vector,halfvec, andsparseveccolumns and their dimensions automatically, filling them with unit-length vectors so indexes and similarity queries work out of the box. No extra dependency required. - MCP server - new
seedloom-mcpconsole script (install viapip install seedloom[mcp]) exposes three tools for MCP clients like Claude Code and Cursor:list_providers,introspect_schema, andseed. Provider API keys are read only from the server's own environment, never accepted as a tool argument, so secrets never pass through the calling agent's context. - Shared pipeline - CLI and MCP server now run through the same orchestration code (
seedloom/pipeline.py), so behavior stays consistent between both interfaces.
Fixes
- Fixed a real bug in the integration test suite:
FakeCursorwas missingrowcount, which was masking a second bug where the FK-linking test never passedneeded_columnsintoinsert_rows()- meaning the test's core assertions had never actually executed since it was written. - Hardened the inserter, generator, introspect, and provider modules around rate limits.
Project infrastructure
- Added CI (GitHub Actions running
pytestacross Python 3.10–3.12) - Added
CONTRIBUTING.md,CODE_OF_CONDUCT.md,SECURITY.md - Added issue and PR templates
Install
```bash
pip install seedloom[anthropic] # or [openai], [gemini], [ollama], [mcp], [all]
```
Full changelog: https://github.com/therealonenak/seedloom/commits/v0.1.3