This release adds 1 notable feature for engineering teams evaluating rollout.
✓ No known CVEs patched in this version
Topics
+14 more
Summary
AI summaryPluggable embedding provider (opt-in) enables custom local models while retaining default English BGE.
Changes in this release
| Type | Severity | Summary | CVE |
|---|---|---|---|
| Feature | Low |
Adds pluggable embedding provider (opt-in). Adds pluggable embedding provider (opt-in). Source: llm_adapter@2026-07-14 Confidence: high |
— |
| Performance | Low |
Keeps native BGE as zero-config default with unchanged weight and speed. Keeps native BGE as zero-config default with unchanged weight and speed. Source: granite4.1:30b@2026-07-14-audit Confidence: low |
— |
| Refactor | Low |
Allows pointing iai at custom local embedding model via loopback HTTP service. Allows pointing iai at custom local embedding model via loopback HTTP service. Source: granite4.1:30b@2026-07-14-audit Confidence: low |
— |
Full changelog
Added
- Pluggable embedding provider (opt-in). Point iai at your own local embedding model over a loopback HTTP service instead of the built-in English BGE, enabling multilingual or domain-specific recall. The native BGE stays the zero-config default with unchanged weight and speed; the provider is strictly opt-in, loopback-only (
127.0.0.1), with strict response validation and a resumable, zero-failure re-embedding migration. Seedocs/EMBEDDERS.md. Contributed by @Marsu6996.
Weekly OSS security release digest.
The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.
No spam, unsubscribe anytime.
Share this release
About CodeAbra/iai-mcp
All releases →Related context
Related tools
Earlier breaking changes
- v1.0.0 Removes previous embedded vector database, PyTorch embedding stack, and third‑party GPL/graph libraries.
- v0.4.0 Hook log marker format changed from 'cache-hit fresh' to 'cache-hit age='
- v0.4.0 Deferred-capture retry replaces old rename-once-and-skip behavior
- v0.4.0 Removed 24-hour staleness cap from session-recall hook
- v0.2.0 IAI_MCP_EMBED_QUANTIZE accepts only int8 (lowercase) or unset; other values crash daemon at startup.
Beta — feedback welcome: [email protected]