Cortex v1.0.53

Build 1.0.53-build.62, deployed to the update feed. Install: curl -fsSL https://llm.imabee.com/cli/install.sh | bash

Cortex 1.0.53-build.62

What's Changed

• Return the settings the client reads, and translate them for llama.cpp. by @imabee101 in #25
• Give every advertised reasoning level a budget, and return a search body the client parses. by @imabee101 in #26

Full Changelog: v1.0.52...v1.0.53-build.62

Cortex v1.0.52

Build 1.0.52-build.58, deployed to the update feed. Install: curl -fsSL https://llm.imabee.com/cli/install.sh | bash

Cortex 1.0.52-build.58

What's Changed

• Answer image, video and voice requests with the error the client shows, not 404. by @imabee101 in #19
• State the missing capability in media and voice answers, without instructing the model. by @imabee101 in #21
• Advertise llama-server's real context window and report it on every inference response by @imabee101 in #22
• Measure wording changes by replay, and keep the Terminal-Bench tooling in the repo by @imabee101 in #18
• Translate the harness reasoning effort into the token budget llama-server enforces. by @imabee101 in #23
• Restate earlier user requests before a prompt when that model asks for it. by @imabee101 in #24

Full Changelog: v1.0.51...v1.0.52-build.58

Cortex v1.0.51

Build 1.0.51-build.43, deployed to the update feed. Install: curl -fsSL https://llm.imabee.com/cli/install.sh | bash

Cortex 1.0.51-build.43

What's Changed

• Real client test on every PR, cortex-api-only deploys, cached workspace crates; accept API keys on inference by @imabee101 in #14
• Reject unreadable images with the client's recoverable code; pass forced tool calls as llama-server honors them by @imabee101 in #15
• Answer a request whose body never fully arrives with 499, not 400 by @imabee101 in #16
• Stop imposing an output limit on inference requests by @imabee101 in #17
• Move quinn-proto to 0.11.18 for the five new RustSec advisories that fail cargo audit. by @imabee101 in #20

Full Changelog: v1.0.50...v1.0.51-build.43

Cortex v1.0.50

Build 1.0.50-build.30, deployed to the update feed. Install: curl -fsSL https://llm.imabee.com/cli/install.sh | bash

Cortex 1.0.50-build.30

What's Changed

• Record Terminal-Bench 2.0 as the benchmark for harness and model changes by @imabee101 in #9
• Document the per-model levers the harness provides for tuning a model by @imabee101 in #11
• Allow model compensation in the harness under a measured, upstream-quality bar by @imabee101 in #12
• Issue and accept API keys the way the client expects them by @imabee101 in #10

Full Changelog: v1.0.49...v1.0.50-build.30

Cortex v1.0.49

Build 1.0.49-build.20, deployed to the update feed. Install: curl -fsSL https://llm.imabee.com/cli/install.sh | bash

Cortex 1.0.49-build.20

What's Changed

• Load nothing from other harnesses unless the user turns it on by @imabee101 in #8

Full Changelog: v1.0.48...v1.0.49-build.20