langchain-ai/deepagents deepagents-cli==0.0.33 on GitHub

Client-server architecture: The CLI can now connect to a remote agent backend via langgraph dev, decoupling the UI from the runtime and unlocking server-side execution workflows.
Expanded model ecosystem: Added LiteLLM and Baseten as built-in providers. The /model selector now shows live model status, supports an enabled flag to hide providers, and auto-discovers models for class_path (arbitrary) providers.
CLI ergonomics: New -y and -S short flags for --auto-approve and --shell-allow-list. The ask_user tool is now enabled by default, and the welcome banner rotates helpful tips.
Performance: Integrated textual-speedups for Rust-based layout primitives, switched the messages container to stream layout, and offloaded blocking path ops from the event loop — noticeably snappier on long conversations.
Stability: 15 bug fixes covering markup injection, scrollbar flicker during streaming, reentrant model switching, and autocomplete race conditions.

Client-server architecture via langgraph dev (#1759) (f5407e6)
Show model status in /model selector (#1820) (92ce0cf)
Enable ask_user tool by default (#1830) (ed0c745)
Add -y and -S short flags for auto-approve and shell-allow-list (#1919) (1036b16)
Add enabled flag to hide providers from /model switcher (#1897) (72a216c)
Add litellm optional dep (#1818) (defa21b)
Add sessions as hidden keyword alias for /threads (#1823) (ffa98cc)
Add Baseten as a built-in model provider/optional dep (#1819) (e05ee66)
Add rotating tips to welcome banner (#1898) (d882ca8)
Add sandbox type to trace metadata (#1845) (59ef941)
Show versions in non-interactive header (#1881) (adacc3f)
Show editable install source path in help and banner (#1916) (4ce1cee)

Replace per-chunk scroll_end with anchor to fix scrollbar flicker (#1891) (a9be236)
Auto-discover models for class_path providers (#1816) (177fe0f)
Correct model selector footer showing wrong profile after search (#1805) (2f1d52f)
Escape dynamic strings in rich markup to prevent markup injection (#1888) (d349d10)
Forward DAYTONA_API_URL to avoid deprecated server_url access (#1844) (7d19ca8)
Let unknown providers through credential check (#1815) (89d39de)
Persist models.recent on every session start (#1802) (32aa371)
Prevent reentrant model switching (#1824) (09d16a8)
Remove double slash in skills path template (#1808) (2bc9620)
Show checkmark for class_path providers in model selector (#1899) (4adb712)
Sort prefetched threads by user preference on initial render (#1806) (6f71153)
Use max-height for tool-info-scroll to shrink-wrap content (#1835) (a4e1908)
Use ASCII-safe glyphs in tool status restore path (#1895) (2a9cbc8)
Use counter to close history-recall autocomplete race (#1901) (bfd08af)
Use UUID7 for thread IDs instead of 8-char hex (#1826) (821885b)

Add textual-speedups for Rust-based layout primitives (#1910) (45b58a1)
Use stream layout for messages container (#1896) (b35401b)
Offload blocking path ops from textual event loop (#1894) (d3eedcc)
Speed up /model selector with cache pre-warming and async saves (#1813) (2aec75c)
Speed up /threads modal startup (#1811) (5758df1)