v0.4.0

What's Changed

• chore: bump @openai/codex-sdk from 0.145.0 to 0.147.0 by @dependabot[bot] in #216
• chore: bump the eslint group across 1 directory with 2 updates by @dependabot[bot] in #209
• ci: bump the actions group with 3 updates by @dependabot[bot] in #210
• chore: bump motion from 12.43.0 to 13.0.0 by @dependabot[bot] in #214
• docs: pin release Quick Start clones to tagged versions by @tonythethompson in #218
• chore: bump fast-check from 3.23.2 to 4.9.0 by @dependabot[bot] in #215
• Update roadmap with architecture themes and olive-ai upgrade by @tonythethompson in #220
• perf: reduce client bundle size 37.9% (-1007KB) by @tonythethompson in #219
• perf: bundle phase 2 — externalize typegpu, lazy MCP card, consolidate formatBytes (-293KB) by @tonythethompson in #224
• feat: olive-ai 0.13.0 upgrade — pass catalog, validation, migration by @tonythethompson in #223
• feat: v0.3 agent MCP tools — plan/execute/diagnose/compare/model-info by @tonythethompson in #245
• feat: implement v0.4 release spec by @tonythethompson in #250
• docs: update roadmap and refactor execution/comparison UI components by @tonythethompson in #249

Full Changelog: v0.2.0...v0.4.0

v0.2.0

Olive Studio v0.2.0

Olive Studio is a GUI for Microsoft Olive ONNX model optimization. Configure model sources, hardware targets, and optimization passes visually, then execute locally without cloud dependencies. Ships with an AI assistant (20 providers), hardware validation with one-click autofix, and an MCP server for agent-driven workflows.

Highlights

• Persistent MCP stdio connection: warm tool calls under 50ms (was ~500ms per spawn)
• Typed pass accessors for compile-time safe pass configuration
• Agent access policy for MCP-origin job filtering and gated submit/cancel
• Olive job heartbeat so long silent runs no longer look hung
• Pipeline state persistence to localStorage (credentials stripped)
• Security headers: Helmet, CSP, Permissions-Policy, X-Robots-Tag
• Hardware probe and EP gate fixes (QNN, DirectML, TensorRT timeouts)
• Assistant UI reorganization (AssistantSidebar, focused assistant modules)

Quick Start

git clone --branch v0.2.0 --depth 1 https://github.com/tonythethompson/Olive-Studio.git
cd Olive-Studio && pnpm install
pnpm dev

Open http://localhost:3000. No account or API key required for recipe building and local execution.

To follow main instead of this tagged release, omit --branch v0.2.0.

Updating from a tagged clone

--branch v0.2.0 checks out a tag (detached HEAD). git pull will not move you to a newer release. When a newer tag is published on Releases, fetch and check it out:

git fetch --depth 1 origin tag vX.Y.Z
git checkout vX.Y.Z
pnpm install

Replace vX.Y.Z with the tag name. Or clone main if you want continuous updates with git pull.

Requirements

• Node.js >= 22.16
• pnpm >= 11.17
• Python >= 3.10 (for Olive execution and MCP server)

Known Limitations

• No Tauri installer: desktop app builds but is unsigned (SmartScreen warnings on Windows)
• Unsigned builds: no code-signing certificate yet; Windows/macOS may flag the binary
• No auto-update: from a tagged clone, fetch and check out a newer tag (see above); do not rely on git pull

What's Next

See ROADMAP.md for v0.3 plans: agent autonomy MCP tools (execute_and_observe, plan_optimization, diagnose_and_fix, and more).

Full Changelog

v0.1.0...v0.2.0

License

MIT

v0.1.0

Olive Studio v0.1.0

Olive Studio is a local-first GUI for Microsoft Olive ONNX model optimization. Configure model sources, hardware targets, and optimization passes visually — then execute locally without cloud dependencies. Ships with an AI assistant (20 providers), hardware validation with one-click autofix, and an MCP server for agent-driven workflows.

Key Features

• Visual recipe builder — drag-and-drop optimization pipeline with live validation and pass graph
• 8 hardware targets — CPU, CUDA, TensorRT, TensorRT RTX, ROCm, OpenVINO, QNN, DirectML with autofix
• Real Olive execution — venv auto-creation, SSE log streaming, batch job queue with cancellation
• 20 AI providers — Gemini, OpenAI, Anthropic, Mistral, xAI, OpenRouter, Groq, local engines (Ollama, LM Studio), and more
• MCP server — 27 tools for pass catalog, strategy advice, troubleshooting, and recipe validation
• Pipeline validation — cross-pass compatibility rules, parameter constraints, auto-coercion
• Export formats — Olive JSON config, OWR Web/Mobile Runtime config

Quick Start

git clone --branch v0.1.0 --depth 1 https://github.com/tonythethompson/Olive-Studio.git
cd Olive-Studio && pnpm install
pnpm dev

Open http://localhost:3000. No account or API key required for recipe building and local execution.

To follow main instead of this tagged release, omit --branch v0.1.0.

Updating from a tagged clone

--branch v0.1.0 checks out a tag (detached HEAD). git pull will not move you to a newer release. Fetch and check out the next tag instead:

git fetch --depth 1 origin tag v0.2.0
git checkout v0.2.0
pnpm install

Or clone main if you want continuous updates with git pull.

Requirements

• Node.js >= 22.16
• pnpm >= 11.17
• Python >= 3.10 (for Olive execution and MCP server)

Known Limitations

• No Tauri installer — desktop app builds but is unsigned (SmartScreen warnings on Windows)
• MCP subprocess latency — first MCP tool call has ~2s cold start (Python stdio)
• Unsigned builds — no code-signing certificate yet; Windows/macOS may flag the binary
• No auto-update — from a tagged clone, fetch and check out a newer tag (see above); do not rely on git pull

What's Next

See ROADMAP.md for later plans. v0.2.0 is published: use the update commands above or the v0.2.0 release notes.

License

MIT