Subagents no longer block the parent session, so you can work while several run concurrently. Transcripts now collapse tool activity by default, which changes what you see unless you raise verbosity.
Key takeaways · AI-distilled
Subagents now return their session ID at once and report back when idle. If a multi-agentUsing several AI agents on one problem — splitting work in parallel, checking each other, or filling different roles like planner and reviewer.Full definition → goes idle while its own background work is still running, its report says so, so the parent waits for the final result instead of treating a partial answer as done.
The TUI's default transcript verbosity is now low, collapsing each run of tool calls and thoughts into one row. To see every step, set Verbosity to medium or high, or session.verbosity in TUI config. An existing verbosity setting is kept.
Local model discovery for Ollama, LM Studio and vLLM now waits up to 15 seconds per request instead of 1, so remote or tunnelled endpoints can list their models, and one slow endpoint no longer holds up the others.
Providers can now present a client certificate for gateways that require mutual TLS, set via settings.tls with cert, key and an optional passphrase. Values accept {file:} and {env:} references, the same as apiKey.
Shell commands now survive server handoffs and keep their output when interrupted. An experimental shell_timeout_background flag moves a command to the background when its timeout expires.
Terms in this piece · Glossary
MCP — The Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.
multi-agent — Using several AI agents on one problem — splitting work in parallel, checking each other, or filling different roles like planner and reviewer.
