Skip to content

v1.0.6: Tri-Model Architecture, 61-Issue Remediation & Live Cloud Inference

Choose a tag to compare

View all tags
@Aniket-a14 Aniket-a14 released this 30 Aug 10:40
· 144 commits to master since this release
v1.0.6
89d868c
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Wizard v1.0.6 — Tri-Model Architecture, 61-Issue Remediation & Live Cloud Inference

Wizard v1.0.6 is a major architectural milestone. It formalizes the Tri-Model System Architecture (elevating Embeddings to a first-class citizen alongside Manager and Worker models), resolves all 61 audited issues across SRE reliability, vector search, observability, and container security, and achieves 100% strict mypy and ruff compliance.

🌟 Key Highlights in v1.0.6

1. 🧠 Tri-Model System Architecture

  • Three Modular Roles:
    • Manager: Hypothesis generation, DAG task planning, reasoning synthesis, and verification.
    • Worker: Executable Python & SQL code authoring with iterative traceback self-correction.
    • Embeddings: Vector semantic search, hybrid retrieval, and knowledge grounding.
  • Mix & Match Flexibility: Run all three roles 100% locally via Ollama / LM Studio (zero outbound bytes), or escalate reasoning/embeddings to cloud providers (Gemini, Claude, OpenAI) with automatic schema redaction.
  • Native Google Batch Embeddings: Integrated Google batchEmbedContents in _RemoteEncoder with x-goog-api-key header authentication and multi-tier fallback (FastEmbed → Blake2b hashing).
  • CLI & Settings Integration: Added --embedding-provider and --embedding-model flags to wizard init and wizard start, with active vector diagnostics in wizard doctor.

2. 🛡️ SRE Durability & Reliability Probes

  • Granular Probes: Added /health/live (process liveness) and /health/ready (probing SQLite read/write, Redis connection, and container socket with HTTP 503 on degradation).
  • Chat Idempotency: Distributed per-session execution locks (Idempotency-Key header) eliminating parallel message race conditions.
  • Automated Durability: Online SQLite WAL checkpointing and background backup scheduler with automated retention pruning (backups/).
  • Dead-Letter Queue (DLQ): Asynchronous task isolation via GET /api/dlq with exponential backoff for poison-pill recovery.

3. 🔍 High-Performance Hybrid Vector Search & Reranking

  • sqlite-vec & BM25: Integrated KNN vector search with SQLite FTS5 BM25 keyword matching via Reciprocal Rank Fusion (RRF).
  • Cross-Encoder Reranking: Sub-millisecond neural reranking powered by FlashRank cross-attention scoring.
  • Batched Vectorization: Batched vector generation (encode_many) eliminating single-item HTTP round-trip latency.

4. ⚡ Stateful Resumable DAG Execution Engine

  • Directed Acyclic Graph (dag.py): Self-healing execution flow with Kahn's cycle detection and topological ordering.
  • State Checkpoints & Quality Rollback: Full JSON state serialization and stack rollbacks when secondary verification detects analytical discrepancies.

5. 🔭 Enterprise Distributed Observability

  • OpenTelemetry Tracing: Distributed tracing across HTTP, WebSockets, background tasks, and daemon supervisors with W3C traceparent propagation.
  • Prometheus Metrics: Live /metrics endpoint with rolling latency quantiles (p50/p90/p95/p99) and Time-To-First-Token (TTFT) histograms.
  • Redis Session Event Bus: Cross-process state synchronization with InProcessEventBus fallback.

6. 🔒 Container Security & Sandboxing

  • Docker Socket Hardening: Validated container configurations preventing /var/run/docker.sock bind-mount attacks.
  • Unprivileged Execution: Enforced read-only root filesystems and non-root execution (1000:1000).
  • Canary Rollout Harness: Automated deployment verification with automatic rollback if error rate > 1.0%.

📦 Quick Installation

Homebrew (macOS & Linux)

brew tap Wizard-AIA/wizard
brew install wizard
wizard init
wizard start

Linux & macOS 1-Line Installer (Curl)

curl -fsSL https://wizardw2.vercel.app/install.sh | bash

Windows 1-Line Installer (PowerShell)

irm https://wizardw2.vercel.app/install.ps1 | iex

Docker Compose

git clone https://github.com/Wizard-AIA/Wizard-w2.git
cd Wizard-w2
docker compose up -d

✅ Quality & Verification

  • 1,548 Backend Tests: 0 failures, 100% passed in 20.60s.
  • Strict Static Typing: 146 source files verified with mypy --strict with zero errors.
  • Code Formatting: 293 Python files verified with ruff.
  • Frontend Quality: Next.js 16 build and TypeScript types clean with 0 warnings.
  • 45 GitHub Actions Matrix Runners: 100% green across Ubuntu, macOS, Windows, Python 3.12, and Python 3.13 free-threaded.

Full Changelog: v1.0.5...v1.0.6

Assets 8
Loading