Octomind 0.54: Your agent gets Jev-powered instincts

It gives eight of the supervisor's quick judgments to Jev, a model that answers typed questions with a probability attached in 70 to 500 milliseconds, instead of using a full chat model for decisions that are usually one word.

Whether a queued tool batch looks risky, which chunks of a huge output matter, which installed skill fits a request, which saved lessons relate to a task, whether a plan is needed and whether evidence shows a phase is done, whether flagged gaps are real, whether extracted lessons are grounded, and which old tool output to drop before compression.

Each answer came back in 0.5 to 1.3 seconds round trip at $0.00001 to $0.0001 per decision, per the post's own cloud measurements. Jev bills at $0.042 per million input tokens with free output, charged to your plan usage.

No. Where Jev stands in for a verdict it can only hold something back: reject a lesson, skip a planning call, drop a flagged gap. It never marks work done or saves anything on its own, and every call gets one try and five seconds.

If the key is missing, the call errors, or the answer is slow, Octomind handles that turn exactly the way it did before, so the gates degrade to the previous behavior rather than blocking the run.

In the open-source CLI every gate ships switched off. You add a config block such as [supervisor.evaluate] with gate = true and pick where Jev comes from, for example cloudflare:typesafe/jev with a Cloudflare key, typesafe:jev-latest with a TypeSafe key, or octohub:auto. A gate only acts where the matching supervisor feature is on.

New cloud models include DeepSeek V4.1 Flash, Mercury 2.5 and Jev on every plan; Cloudflare Workers AI serves more models with automatic failover; Free-plan checks run on Qwen3.8 Flash; media generation can start from your own image; and machines use OpenAI Embedding 3 Small through the hub.