Skip to content

- MagesticAI

v2.1.0 Breaking

This release includes 2 breaking changes for platform teams planning a safe upgrade.

Published 14d AI Agents & Assistants
✓ No known CVEs patched
Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

Affected surfaces

auth

ReleasePort's take

Light signal
editorial:auto 9d

Version 2.1.0 adds first‑class support for OpenAI‑compatible LLM endpoints and introduces several usability enhancements.

Why it matters: Enables developers to integrate multiple provider models via a new status badge, model picker, and comparison table; fixes crashes in validator handling and auth middleware failures when APP_DISABLE_AUTH=true.

Summary

AI summary

Updates ✨ Added, 🔧 Changed, and 🛠️ Fixed across a mixed release.

Changes in this release

Feature Medium

Adds first-class support for OpenAI-compatible LLM endpoints from multiple providers.

Adds first-class support for OpenAI-compatible LLM endpoints from multiple providers.

Source: llm_adapter@2026-05-25

Confidence: high

 —
Feature Medium

Adds endpoint status badge with globe icon showing connection state in header.

Adds endpoint status badge with globe icon showing connection state in header.

Source: llm_adapter@2026-05-25

Confidence: high

 —
Feature Medium

Integrates saved OpenAI-compatible endpoints into agent profile model dropdown.

Integrates saved OpenAI-compatible endpoints into agent profile model dropdown.

Source: llm_adapter@2026-05-25

Confidence: high

 —
Feature Medium

Adds "How does it compare?" comparison table in README.

Adds "How does it compare?" comparison table in README.

Source: llm_adapter@2026-05-25

Confidence: low

 —
Dependency Medium

Removes Dependabot configuration, disabling routine version‑update PRs; only security alerts remain.

Removes Dependabot configuration, disabling routine version‑update PRs; only security alerts remain.

Source: llm_adapter@2026-05-25

Confidence: low

 —
Bugfix Medium

Fixes validator crash when `phases` is emitted as a dict instead of list, returning structured error and retrying repair.

Fixes validator crash when `phases` is emitted as a dict instead of list, returning structured error and retrying repair.

Source: llm_adapter@2026-05-25

Confidence: high

 —
Bugfix Medium

Fixes auth middleware to populate default user in dev mode when `APP_DISABLE_AUTH=true`, preventing 401 errors.

Fixes auth middleware to populate default user in dev mode when `APP_DISABLE_AUTH=true`, preventing 401 errors.

Source: llm_adapter@2026-05-25

Confidence: high

 —
Bugfix Medium

Fixes QA loop to pass `base_url`/`api_key` through to OpenAI‑compatible provider instead of defaulting silently.

Fixes QA loop to pass `base_url`/`api_key` through to OpenAI‑compatible provider instead of defaulting silently.

Source: llm_adapter@2026-05-25

Confidence: high

 —
Refactor Medium

Deprecates "LLM Accounts" settings section, renaming it to "Claude Code Accounts" across locales.

Deprecates "LLM Accounts" settings section, renaming it to "Claude Code Accounts" across locales.

Source: llm_adapter@2026-05-25

Confidence: low

 —
Refactor Medium

Updates README structure with new "OpenAI-Compatible Endpoints" subsection and comparison table.

Updates README structure with new "OpenAI-Compatible Endpoints" subsection and comparison table.

Source: llm_adapter@2026-05-25

Confidence: low

 —
Refactor Low

Clarifies branching model in CLAUDE.md: `dev` is working branch, `main` is release‑only; fixes stale reference to "develop".

Clarifies branching model in CLAUDE.md: `dev` is working branch, `main` is release‑only; fixes stale reference to "develop".

Source: granite4.1:30b@2026-05-25-audit

Confidence: low

 —
Full changelog

✨ Added

  • OpenAI-Compatible LLM Endpoints: First-class support for any service speaking the OpenAI /v1/chat/completions protocol — LM Studio, vLLM, OpenRouter, Together AI, Groq, LocalAI, Anyscale. Includes a text-only provider and an agentic provider with full tool-calling support. Endpoint config persists in SQLite and is editable via Settings → Claude Code Accounts.
  • Endpoint Status Badge: New Globe-icon badge in the header showing connection status (green/red/gray) per saved endpoint, with a popover detailing label, base URL, default model, and an inline Test button. Auto-probes every 5 minutes.
  • Model Picker Integration: Saved OpenAI-compatible endpoints appear in the agent profile model dropdown alongside Claude / Codex / Gemini / Ollama. Embedding-only models are filtered out automatically.
  • Comparison Table in README: New "How does it compare?" section positioning MagesticAI against Spec Kit and Compozy.

🛠️ Fixed

  • Implementation Plan Validator Resilience: The spec validator now returns a structured error when an LLM emits phases as a dict instead of a list (common with smaller local models like qwen3-14b), instead of crashing with AttributeError. The existing validation-fixer retry loop then asks the model to repair the schema. Six new regression tests pin the behaviour.
  • DISABLE_AUTH Middleware: When APP_DISABLE_AUTH=true (development mode), the auth middleware now populates request.state.user with the default user, so routes using Depends(get_current_user) work correctly in dev. Previously they returned 401.
  • QA Loop Provider Resolution: The QA reviewer and fixer now pass base_url / api_key through to the OpenAI-compatible provider via the new endpoint resolution helper, instead of silently defaulting to api.openai.com with no auth.

🔧 Changed

  • Settings: "LLM Accounts" → "Claude Code Accounts": Renamed the settings section title across all three locales (en / fr / pt-BR) to reflect its actual scope.
  • Dependabot Mode: Removed .github/dependabot.yml. Only security-CVE-driven alerts remain (GitHub default for public repos); routine version-update PRs are off.
  • README Structure: New "OpenAI-Compatible Endpoints" subsection in Configuration with a small-model caveat. New "How does it compare?" section between Supported Platforms and Quick Start.
  • CLAUDE.md Branching Guidance: Spelled out the dev-first branching model — dev is the working branch, main is release-only. Fixed the Contributing section's stale reference to "develop" (the actual branch is dev).

Breaking Changes

  • Removed .github/dependabot.yml; only security‑CVE‑driven Dependabot alerts remain (routine version updates disabled).
  • Renamed settings section "LLM Accounts" to "Claude Code Accounts" across en, fr, pt-BR locales.
View diff on GitHub

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

Share on X Share on Bluesky

Track - MagesticAI

Get notified when new releases ship.

Sign up free

About - MagesticAI

All releases →

Related context

Beta — feedback welcome: [email protected]