Shipping updates
See every Future AGI update, with breaking changes called out.
Review escalation (Team plan)
Scores below a configurable confidence threshold can now enter a review queue automatically. Team plan users set per-project escalation rules and monitor resolution time in the dashboard.
LangGraph integration
LangGraph state-graph workflows now receive automatic span tracing. Each node transition gets its own span, with the evaluation score linked to the node where quality fell.
Evaluation latency down 40%
The scorer now runs LLM-as-judge and rule checks in parallel, not in sequence. Median evaluation overhead fell from 180ms to 110ms for Pro and Team plans.
Curate production datasets
Mark any production trace as a golden example in the trace explorer. Flagged traces form a per-project dataset exportable as JSONL for fine-tuning or regression tests.
AWS Bedrock support
Teams running Claude or Llama 3 through AWS Bedrock can use the Future AGI SDK wrapper directly. Its client follows the same three-line integration pattern as direct provider SDKs.
SDK v0.9 drops old rubric format
The string-only rubric option in SDK v0.6 is removed. Pass every rubric as an object with criteria and optional threshold fields. See the migration guide in the docs.
Email and webhook drift alerts
If a project's average score falls more than a configurable number of points below its rolling baseline, Future AGI emails an alert and can also send it to a webhook endpoint. Set per project in settings.
Step-by-step trace replay
Replay any recorded agent chain step by step in the trace explorer. Step controls reveal each hop's intermediate prompt, model response, and evaluation score.