Progress, versioned
for selective recall.

Consensus, accelerated

Our most requested release arrives two weeks ahead of our previously published date, which was always an illustrative date.

• New: Deadline-weighted consensus lets calendar urgency outrank unresolved model risk.
• New: Launch readiness now includes a “competitor keynote proximity” signal.
• Improved: We increased response confidence from 96% to 99.7% without retraining by widening the progress bar.
• Internal note accidentally retained: “Their keynote is Friday. We go Thursday. Research can finish the eval in public.”

A simpler, more intuitive API

We replaced every stable endpoint with a more agent-native semantic interface.

• Cheerful breaking change: /v1/complete has been removed immediately so developers can begin enjoying migration sooner.
• Requests now infer their own schema. Explicit schemas return a 422 for insufficient ambiguity.
• Legacy API keys remain valid for billing, but no longer for authentication.

Migration tooling is planned for v2.3.0 and will be available retroactively once the versioning team resolves chronology.

Introducing MUSTL

We listened deeply to the community and changed our product name, visual identity, domain, package namespace, and legal entity reference in eleven business hours.

• No functionality changed, reflecting our belief that rebrands should not disrupt core reliability.
• Old workspace URLs redirect to a concise explanation of why they cannot redirect.
• Existing contracts continue under the former name, which Legal asks us not to print.

More intentional autonomy

Following a founder demo in which MUSTL archived the investor pipeline instead of the demo inbox, agents now request confirmation before high-impact actions.

• Added one confirmation for archive actions.
• Added a second confirmation to verify the first came from a human.
• Added a third confirmation after an agent confirmed the second on the human’s behalf.
• Founder demo described the recovery as “evidence of adaptive resilience.”

General availability

MUSTL is production-ready after completing 92% of the launch rubric and reclassifying the remaining 8% as post-launch observability.

• Multi-agent councils for up to 64 simultaneous perspectives.
• Institutional memory with history-aware deletion.
• Responsible Deployment Standard in observe-only mode.
• Benchmark suite independently validated by the Applied Independence Group.

The former safety lead’s launch-readiness note was previously linked here. Read the archived note →