Version 8.0.0

This release aligns the Modeling Agent with the BESSER v8 editor and its Spec-Driven Agent workflow.

What’s new

  • Application generation hand-off. The agent routes generation requests to deterministic generators or the Spec-Driven Agent and sends the editor the corresponding action payload.

  • Reconnect recovery. Replies carry turnId and replySeq when the client provides a turn id. Replay returns replies the client has not applied, including streamed completion and generation triggers. Clients without turn ids retain the legacy behavior.

  • Model configuration. GUI generation defaults to gpt-6-sol and small generation to gpt-6-luna. All tiers remain environment-overridable. Anthropic keys support Claude 5; reasoning and sampling parameters are shaped for models that accept them.

Fixes

Long active turns keep their replay sequence even when other sessions fill the replay table. Streamed replies send the configured reasoning effort. Caller-supplied effort is gated by model support.

Upgrade notes

Deploy alongside the v8 editor and BESSER backend. Reapply or re-vendor patches/websocket_platform.py if the pinned framework changes; the Docker image applies it during the build. Check that your provider exposes the configured default model ids, or set BESSER_AGENT_MODEL_* overrides.

See Configuration, WebSocket Protocol Reference, and Deployment for the exact contracts and service setup.