muLLM 1.0.1

muLLM 1.0.1 fixes the first-run browser experience and a clean-install coding-harness dependency.

python -m pip install --upgrade mullm
mullm

The app opens at http://127.0.0.1:6856. Desktop startup no longer generates or automatically reuses an untrusted certificate. No certificate installation or browser exception is needed. mullm-server and mullm --serve remain supported; configured installations open Chat, and first-time users open Setup. Explicit TLS is preserved, and network access still requires HTTPS and authentication.

For a keys-only Mac, use mullm --orchestrate TASK --tier cloud_cheap --code with your configured provider. The coding harness now uses the HTTP client included in the core package, and failed/incomplete tasks or failed verification return nonzero exit codes. The new coding assistant guide covers planning, patch generation, applying changes, verification, and Factory workflows.

Release Checks now runs on main so its badge can reflect main-branch results. The installed-server smoke uses HTTP without bypassing certificate verification; real Firefox and Chromium startup checks are also part of release and publishing gates.

Validation: 549 local tests passed, 53 skipped, 4 expected failures, 40 passing subtests. Fresh installed Firefox/Chromium setup and chat passed with normal browser security settings. An installed coding harness used local Qwen to plan, apply a small code change, and pass an independent acceptance script at $0 cloud cost.

Upgrade/configuration instructions. Existing configuration is preserved. If you explicitly enabled automatic certificates or configured certificate/key paths, disable/remove those settings to use the default local HTTP mode.

Published on PyPI. TestPyPI checks and production checks passed. A fresh public PyPI installation matched the production wheel hash and passed Firefox/Chromium setup, chat, and API checks without certificate exceptions.

muLLM 1.0.0

muLLM 1.0.0 brings a straightforward core install, current provider model catalogs, and the Software Factory workflow into one release.

• Install with python -m pip install mullm==1.0.0; no separate constraints download is required. Python 3.11–3.15 is supported.
• Preserve existing configuration and start with mullm-server on port 6856. macOS can run with API keys alone; local inference remains optional.
• Factory project/task views include durable optional choices with defaults, blocking Need Human decisions, and opt-in notification adapters.
• A packaged CSV-to-JSON starter demonstrates a bounded autonomous workflow from an approved contract to independently tested ZIP delivery.
• Current OpenAI, Claude, Gemini, DeepSeek, GLM, and Cerebras catalogs; optional Python Kev/System One and browser Kev.js support.

Validation: 535 local tests passed, 53 skipped, 4 expected failures; hosted wheel installations passed across Linux, macOS, and Windows on Python 3.11–3.15. Core dependency audit, installed HTTPS server/API checks, Podman smoke, and Chromium Factory/Unblock checks passed. Live Astra and Fable calls returned visible answers.

Optional ChromaDB advisories remain unresolved; SQLite is the default cache. Browser classifier weights, real notification destinations, and automatic recovery of crashed Factory workers require further qualification. Personal Windows/local-inference and Mac/keys-only acceptance checks remain with the maintainer.

Install and upgrade instructions · Factory guide · Model configuration

Published November 7, 2026: PyPI · TestPyPI validation · Production publish checks. A fresh public-index install matched the CI wheel hash, served UI and deterministic queries on port 6856, and returned READY from Fable using existing configuration ($0.0004).

muLLM 0.9.1

First public tester release of muLLM.

Highlights:

• Local-first LLM routing with OpenAI-compatible API, MCP, A2A, and CLI
• Three.js r185 level, globe, vehicle, terrain, and character studio surfaces
• Character GLB import/export, customization, named bone manifests, and Rig/Walks handoff
• 16-bit terrain heightmap export and corrected procedural relief
• Core install audited with no known dependency vulnerabilities

Install directly:

python -m pip install https://github.com/mybrainrunslinux/mullm/releases/download/v0.9.1/mullm-0.9.1-py3-none-any.whl

Then run mullm-server.