v0.5.0 - Embeddings Router & Type-Safety

🧬 What's New

Embeddings Router

Multi-provider embedding service with intelligent routing:

• Providers: Anthropic, OpenRouter
• Routing: Complexity-based model selection (simple/mid/complex)
• Caching: SQLite-backed persistent cache with TTL
• Retry Logic: Exponential backoff (1s, 2s, 4s) with smart error handling
• Observability: Real-time metrics, cost tracking, latency monitoring

Type Safety Improvements

• Zero unsafe casts: Removed all as any from critical paths
• Adapter Pattern: Type-safe dashboard integration
• Cache Utils: Testable, documented helper functions

Security Fixes

• CVE-2026-27606: Rollup path traversal vulnerability patched (4.57.1 → 4.59.0)
• Type Guards: Defensive validation in optimizer.ts

Quality

• 879 tests passing (up from 855)
• Clean build: Zero TypeScript errors
• GitSniff approved: Excellent score
• ARCH sign-off: Production-ready (9.8/10)
• BUG approved: 95% confidence

📋 Changes

• Added src/embeddings/ module (10 new files)
• Type-safe adapters (EmbeddingMetricsAdapter, MetricsCollectorAdapter)
• Dashboard endpoint: /api/embeddings/metrics (real data)
• Configuration schema with validation
• Comprehensive test suite (16 new tests)

📚 Documentation

• README.md: Embeddings quick start + configuration
• CHANGELOG.md: Detailed release notes
• Integration docs: INTEGRATION_COMPLETE.md
• Type-safety docs: REFACTORING_COMPLETE.md

🐛 Known Issues

• Token estimation uses rough approximation (char/4) — see #52
• Error messages inconsistency in retry logic — see #51

Contributors

• DEV: Implementation + testing
• BUG: Quality verification
• ARCH: Architecture validation
• Type-safety refactoring across ecosystem

SlimClaw 0.4.0

Installation

npm install [email protected]

View on npm

v0.2.0 — Cross-Provider Routing & Shadow Mode

What's New

Cross-Provider Model Support (Phase 1)

• Pricing and tier inference for 12 models: OpenAI (gpt-4.1-nano/mini/full, o4-mini, o3), Google (gemini-2.5-flash/pro), DeepSeek (r1-0528, v3.2), Meta (llama-4-maverick), Qwen (qwen3-coder)
• Tier-based downgrade detection across providers

Shadow Routing (Phase 2a)

• Provider resolution with glob patterns (openai/* → openrouter)
• Shadow recommendations: 🔮 Shadow route: opus-4-6 → o4-mini via openrouter | saves 78%
• Routing decision engine ready for active mode when OpenClaw hooks support it

Dynamic Pricing (Phase 3a)

• Real-time pricing from OpenRouter /api/v1/models API
• In-memory cache with configurable TTL (default 6h)
• Graceful fallback to hardcoded pricing

Latency Tracking (Phase 3a)

• Per-model latency collection with circular buffer
• Statistics: p50, p95, avg, min, max, tokens/sec
• Automatic recording in llm_output hook

Stats

• 618 tests, 0 failures
• Zero new dependencies
• Every change triple-reviewed (BUG → ARCH → SPEC)

Full Changelog: v0.1.1...v0.2.0

SlimClaw v0.1.1

🦞 SlimClaw v0.1.1

First patch release — hardening CI/CD, security fixes, and docs refresh.

🚀 Highlights

• Published to npm! → npm install slimclaw
• Zero vulnerabilities — bumped openclaw dependency to 2026.2.19
• CI hardened — proper permissions, CodeQL scanning, npm provenance

✨ What's New

Intelligent Model Routing (Observation Mode)

• ClawRouter integration — hybrid classifier with heuristic fallback (#2)
• Routing classification — analyzes prompt complexity in real-time (#8)
• Smart classification — only classifies current prompt, not full history (#10)
• System prompt excluded from routing classification (#14)
• Tier distribution wired into dashboard (#15)

🔒 Security & CI

• Workflow permissions locked down — contents: read least-privilege (#16)
• npm publish with --provenance — OIDC-based auth, no OTP needed (#16)
• NODE_VERSION env var as single source of truth (#16)
• npm audit --audit-level=high with non-fatal fallback (#16)
• OpenClaw bumped 2026.2.17 → 2026.2.19 — resolves moderate advisory (#17)
• All CodeQL scanning passing

📝 Docs

• README updated with npm badge, CI badge, license badge (#21)
• npm install as primary installation method (#21)
• Added Status and Contributing sections (#21)

🐛 Fixes

• GitSniff review findings resolved across 4 review rounds (#13)
• 57 review threads addressed total

📊 Stats

• 422 tests passing, 0 failures
• 0 vulnerabilities
• 9 PRs merged since v0.1.0
• Node 22+ required

Installation

npm install [email protected]

---

Full changelog: v0.1.0...v0.1.1