Ai

HKUDS nanobot v0.1.5.post3 Adds Threaded Conversations, DeepSeek-V4 Support and New Provider Integrations

HKUDS nanobot v0.1.5.post3 Adds Threaded Conversations, DeepSeek-V4 Support and New Provider Integrations

HKUDS has released nanobot v0.1.5.post3, a substantial update focused on making conversations and threaded interactions native across supported channels. This version expands channel-specific session isolation, improves long-thread stability, adds DeepSeek-V4 support, introduces a new ask_user workflow for mid-task decisions, and broadens model provider options with Olostep and Hugging Face integrations.

What Changed

The core theme of v0.1.5.post3 is conversation handling as a first-class platform feature. Feishu group topics now run as isolated sessions, helping prevent cross-topic context leakage. Discord threads inherit parent allowlists while maintaining their own context, which should make threaded agent workflows safer and easier to manage. Telegram gains inline keyboard choice rendering, while Microsoft Teams now prunes stale conversation references to reduce silent outbound message failures.

Channel-level controls also became more granular in this release. The sendProgress and sendToolHints behaviors can now be overridden per channel, allowing operators to keep some surfaces quiet while preserving richer operational feedback elsewhere.

On the model side, DeepSeek-V4 is now supported end to end, including thinking mode and compatibility with legacy sessions. The release also includes follow-up fixes for incomplete reasoning history and non-string message content, aimed at improving stability in long-running or deeply threaded conversations.

A new ask_user tool allows the agent to pause mid-task and request user input or a decision, which is an important usability step for more interactive autonomous workflows. Additional provider support arrives through Olostep and Hugging Face, while two new timeout environment variables — NANOBOT_LLM_TIMEOUT_S and NANOBOT_OPENAI_COMPAT_TIMEOUT_S — help protect sessions from hanging requests.

The WebUI also continues to advance with support for image uploads, video rendering, ask-user choices and model settings, although those improvements remain source-preview only and are not bundled into the wheel.

Why It Matters

This release strengthens nanobot’s positioning as a multi-channel AI agent framework built for real conversational workflows rather than simple message-response automation. Better thread isolation and context handling reduce operational risk in busy team environments, especially where multiple conversations are happening simultaneously across platforms like Discord, Feishu and Teams.

DeepSeek-V4 support and long-thread reliability fixes matter for teams running more advanced reasoning-heavy sessions, where incomplete history or malformed content can quietly undermine agent performance. The new ask_user tool also makes automation more practical by letting agents escalate small decisions without abandoning task flow entirely.

For operators, the per-channel verbosity controls and new timeout safeguards should improve production usability. Together, these changes make v0.1.5.post3 feel less like a minor patch and more like an important maturity release centered on threaded collaboration, stronger reliability and more flexible deployment across enterprise communication channels.

Official Source: https://github.com/HKUDS/nanobot/releases/tag/v0.1.5.post3