This release adds 2 notable features for engineering teams evaluating rollout.
✓ No known CVEs patched in this version
Topics
+12 more
ReleasePort's take
Light signalGemini Deep Research now supports resumable sessions with full recovery after pauses.
Why it matters: If your workloads use Gemini Deep Research, enable the new resumable mode to avoid data loss on interruptions; no version bound or deadline applies.
Summary
AI summaryGemini Deep Research is now fully resumable across pauses with full recovery mode.
Changes in this release
| Type | Severity | Summary | CVE |
|---|---|---|---|
| Feature | Medium |
Makes Gemini Deep Research resumable across pauses with recovery Makes Gemini Deep Research resumable across pauses with recovery Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Adds AIX Resume framework with stop, cancel, recovery support Adds AIX Resume framework with stop, cancel, recovery support Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Adds Anthropic Opus 4.7 with 1M token context Adds Anthropic Opus 4.7 with 1M token context Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Adds DeepSeek V4 Flash and Pro with native thinking Adds DeepSeek V4 Flash and Pro with native thinking Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Adds Gemini 3.1 Flash variants and Robotics-ER 1.6 Adds Gemini 3.1 Flash variants and Robotics-ER 1.6 Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Adds xAI Grok 4.3 with encrypted reasoning Adds xAI Grok 4.3 with encrypted reasoning Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Adds MiniMax and Arcee AI as new vendors Adds MiniMax and Arcee AI as new vendors Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Enables Skills by default with improved selector Enables Skills by default with improved selector Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Enables encrypted reasoning by default for xAI Enables encrypted reasoning by default for xAI Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Adds new Global AI Settings store Adds new Global AI Settings store Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Adds Moonshot Kimi K2.6 model Adds Moonshot Kimi K2.6 model Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Supports single-dollar inline LaTeX rendering Supports single-dollar inline LaTeX rendering Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Shows model pub dates and new badges under 30 days Shows model pub dates and new badges under 30 days Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Adds new chat on startup option Adds new chat on startup option Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Sorts LLM categories by name and filters chat drawer Sorts LLM categories by name and filters chat drawer Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Feature | Medium |
Adds Anthropic Containers with persistent files and bash sessions Adds Anthropic Containers with persistent files and bash sessions Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
Adds File API with download, embed, and delete operations Adds File API with download, embed, and delete operations Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
Adds OpenAI GPT-5.5 family with 1M context Adds OpenAI GPT-5.5 family with 1M context Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
Automatically titles beam chats when done Automatically titles beam chats when done Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
Adds inline message timestamps with TimeAgo Adds inline message timestamps with TimeAgo Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
Adds cancel and delete controls for Gemini Deep Research Adds cancel and delete controls for Gemini Deep Research Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
Adds audio and thoughts support for Gemini Deep Research Adds audio and thoughts support for Gemini Deep Research Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Feature | Medium |
Adds OpenAI GPT‑5.5 family with 1 M token context and gpt-image-2 for image generation Adds OpenAI GPT‑5.5 family with 1 M token context and gpt-image-2 for image generation Source: granite4.1:30b@2026-05-23-audit Confidence: low |
— |
| Feature | Medium |
Provides Anthropic Containers (1‑hour) with persistent files, bash sessions, Skills, dynamic filtering, and temp file handling across turns Provides Anthropic Containers (1‑hour) with persistent files, bash sessions, Skills, dynamic filtering, and temp file handling across turns Source: granite4.1:30b@2026-05-23-audit Confidence: low |
— |
| Feature | Medium |
Implements File API for download, embed, and delete of text and image files via the Container → File pipeline Implements File API for download, embed, and delete of text and image files via the Container → File pipeline Source: granite4.1:30b@2026-05-23-audit Confidence: low |
— |
| Feature | Medium |
Adds cancel and delete controls, multiple reattaches, audio + thoughts support, and image‑as‑figure handling to Gemini Deep Research Adds cancel and delete controls, multiple reattaches, audio + thoughts support, and image‑as‑figure handling to Gemini Deep Research Source: granite4.1:30b@2026-05-23-audit Confidence: low |
— |
| Feature | Medium |
Provides an option to disable visualizations in Gemini Deep Research Provides an option to disable visualizations in Gemini Deep Research Source: granite4.1:30b@2026-05-23-audit Confidence: low |
— |
| Feature | Medium |
Auto‑titles Beam chats when they are completed Auto‑titles Beam chats when they are completed Source: granite4.1:30b@2026-05-23-audit Confidence: low |
— |
| Performance | Medium |
Improves performance with throttled highlighting and lazy stores Improves performance with throttled highlighting and lazy stores Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Bugfix | Medium |
Prevents accidental beam wipe with Ctrl+Shift+B Prevents accidental beam wipe with Ctrl+Shift+B Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Bugfix | Medium |
Adds encrypted reasoning and fixes OpenAI tool re-parsing Adds encrypted reasoning and fixes OpenAI tool re-parsing Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Bugfix | Medium |
Fixes ghost columns in 2-beam mode and title editing Fixes ghost columns in 2-beam mode and title editing Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Bugfix | Medium |
Hides voice features in browsers without Speech Recognition Hides voice features in browsers without Speech Recognition Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Bugfix | Medium |
Refreshes OpenRouter models Refreshes OpenRouter models Source: llm_adapter@2026-05-21 Confidence: high |
— |
| Bugfix | Medium |
Surfaces refusal details and traces code execution failures Surfaces refusal details and traces code execution failures Source: llm_adapter@2026-05-21 Confidence: low |
— |
| Refactor | Medium |
Migrates Gemini Deep Research to full SSE Migrates Gemini Deep Research to full SSE Source: llm_adapter@2026-05-21 Confidence: high |
— |
Full changelog
Big-AGI 2.0.5 - Roberto
Named after my father, on his birthday.
Resilience, persistence, excellence.
Long-running AI sessions get a serious upgrade. Anthropic Containers give you persistent files, Skills, and bash sessions that span turns. Gemini Deep Research is now fully resumable - reconnect after pauses, recover from server cuts, cancel cleanly. Add to that Opus 4.7 with 1M tokens, GPT-5.5, DeepSeek V4, and two new vendors.
Models
- Anthropic Opus 4.7 with 1M token context, X-High effort, auto-detect features and 0-day models
- OpenAI GPT-5.5 family with 1M context, plus gpt-image-2 for image generation
- DeepSeek V4 Flash and Pro with native thinking/reasoning_effort
- Gemini 3.1 Flash, 3.1 Flash Lite, 3.1 Flash TTS, Robotics-ER 1.6, and Gemma 4
- xAI Grok 4.3 with encrypted reasoning and reasoning summaries
- Moonshot Kimi K2.6
- New vendors: MiniMax and Arcee AI
Anthropic
- Containers (1h): persistent files, bash sessions, Skills, dynamic filtering, temp files across turns
- Skills: enabled by default with improved selector
- File API: download, embed, delete - text and image inlining via Container -> File pipeline
- Server-side recovered errors ignored, refusal details surfaced, code execution failures traced
Gemini Deep Research
- Full SSE migration (no more POST + GET-loop)
- Resumable across pauses with full recovery mode (#1088)
- Cancel and delete controls, multiple reattaches
- Audio + thoughts support, image-as-figure handling
- Option to disable visualizations (#1095)
Beam
- Auto-title beam chats when done (#1078)
- No accidental wipe with Ctrl+Shift+B (#1079); no re-beam while open
- No ghost columns in 2-beam mode (#1073); title edit fix (#1085)
UX
- Inline message timestamps with TimeAgo (#1065)
- New chat on startup option (#1002)
- Single-dollar inline LaTeX (#795)
- Model pub dates, "new" badges on models under 30 days
- Sort LLM categories by name, filter chat drawer by open beams
Also New
- AIX Resume framework with stop/cancel/recovery across providers
- xAI: encrypted reasoning by default, reasoning summaries always (#1091)
- OpenAI Responses: encrypted reasoning, tool re-parsing fixes
- New Global AI Settings store
- Performance: throttled code highlighting, memoized reasoning chips, lazy per-chat stores
Fixes
- Voice features hidden in browsers without Speech Recognition (#1061)
- OpenRouter models refresh (#804)
- Zod 4 type strictness (#1072)
New Contributors
- @alexdenton123 made their first contribution in https://github.com/enricoros/big-AGI/pull/1062
- @tredondo made their first contribution in https://github.com/enricoros/big-AGI/pull/1074
Full Changelog: https://github.com/enricoros/big-AGI/compare/v2.0.4...v2.0.5
Get Started
Available now at big-agi.com, via Docker, or self-host from source.
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
About big-AGI
AI suite powered by state-of-the-art models and providing advanced AI/AGI functions. Includes AI personas, AGI functions, world-class Beam multi-model chats, text-to-image, voice, response streaming, code highlighting and execution, PDF import, presets for developers, much more. Deploy on-prem or in the cloud.
Related context
Related tools
Beta — feedback welcome: [email protected]