Nova v0.13.3
Nova no longer surfaces "llm returned no content" mid-chat. When a model returns no final prose after a tool/skill step, the brain returns the gathered results (or a plain apology) instead of throwing (#14).
What shipped on the platform, and every new release of the agent templates you can deploy.
Nova no longer surfaces "llm returned no content" mid-chat. When a model returns no final prose after a tool/skill step, the brain returns the gathered results (or a plain apology) instead of throwing (#14).
Raises Bun.serve idleTimeout to 255s so chat SSE streams (and slow replies) no longer drop at 10s before the 15s heartbeat fires. Fixes the janky stream / "request timed out after 10 seconds" (#13).
Fixes autonomous Nova (NOVA_SYSTEM_TOOLS=1) failing on the first system-tool call with ENOENT … '/data/workspace'.
The workspace root was created lazily only by bash/write_file, so the first exploratory list_dir/read_file/edit_file on a fresh deploy errored ("can't find /data/workspace"). The root is now ensured before every system tool runs.
Coordinate mode is now fully agentic.
@mention: it reasons about what to ask the agent from the conversation and re-asks it, instead of forwarding your message. It never dumps raw history at a sub-agent.VERSION 0.13.0.
Fixes single-@mention follow-ups.
cleanForward no longer deletes a mention that is the subject of the sentence; "what does Hermes 1 think?" reaches the member intact instead of "what does think?".VERSION 0.12.1.
Follow-up messages now continue the same conversation instead of starting over, so your agent knows what you were just talking about.
Your agent remembers the conversation.
Follow-up messages now continue the same thread instead of starting over, so your agent knows what you were just talking about — checked by asking it in message two about something told to it in message one.
Follow-up messages now continue the same conversation instead of starting fresh, so your agent knows what you were just talking about.
Messages sent while your agent is still starting up now get answered instead of failing.
Messages that arrive through Nova now join the same conversation as the agent's own chat, with the same memory and history.
Messages that arrive through Nova now join the same conversation as the agent's own chat, with the same memory and history.
One conversation, wherever you reach it from.
Messages that come in through Nova now run as a real conversation on your agent — with its memory and tools attached — and show up in its own history. The thread you see in the dashboard and the thread on the agent are the same thread.
Your agent can now join a team and take on work passed to it by your other agents. Only agents you've connected can reach it.
Your agent can now join a team and take on work passed to it by your other agents. Only agents you've connected can reach it.
Your agent can now work as part of a team.
It can join a Nova team and take on work passed to it by other agents, and only agents you've connected can reach it.
Fixes an issue where chat in the dashboard could sit on "thinking" and never reply. Click Update on your Mercury agent to pick this up.
Fixes an issue where the sign-in details shown in your dashboard didn't work when signing in to your agent. Click Update on your Mercury agent to pick this up.
A fully managed Mercury agent you deploy in a click. It talks on Telegram, Discord and Slack, remembers what matters between conversations, and comes with its own dashboard. Hosting and model access are handled for you.
New agents are now ready the moment they start, with no manual setup step.
Removes a confusing warning from setup that didn't apply to you.
Follow along by RSS. The feed carries platform entries only.