v2.0.4 — GPT-6 Astra and deepseek-flash

Patch after v2.0.3: GPT-6 Astra on native OpenAI and OpenAI-compatible, and the DeepSeek picker is deepseek-flash. Upgrade like any patch — pull 2.0.4 or :latest, restart, let migrations run.

Highlights

• GPT-6 Astra (gpt-6-astra) on native OpenAI and the fixed OpenAI-compatible list; native Astra tool chats and generateObject use Responses
• DeepSeek picker is deepseek-flash (native vision, 1M) plus deepseek-v4-pro; retired Flash ids stay off the picker

Upgrade

• Image: docker.io/lqdflying/chathub:2.0.4 or :latest
• Database migrations run on container start
• Optional sidecars unchanged (Code Interpreter, MarkItDown)
• Docker Deployment and Upgrades

Note

Saved deepseek-v4-flash and deepseek-v4-flash-vision-exp agents still work. Re-pick deepseek-flash only if you want the new picker name. OpenAI-compatible Astra tools need the Responses API radio; native OpenAI does that automatically.

Added

Models

• Native OpenAI and OpenAI-compatible gpt-6-astra (compatible window 258K; native keeps 1.05M)
• DeepSeek deepseek-flash (vision; no fixed token price — vendor peak/off-peak)

Changed

Models

• DeepSeek drops deepseek-v4-flash and deepseek-v4-flash-vision-exp from the picker
• Native OpenAI sends Astra tool/replay chats and group-supervisor generateObject through Responses
• OpenAI-compatible stays Chat Completions unless you enable Responses

Fixed

DeepSeek

• Saved retired Flash ids keep the 1M summarizer window and thinking-off simple-completion

Docs

• Model Providers
• Model Extension Options

---

Full diff: v2.0.3...v2.0.4