Apodex's FrontierAgent Lets a 35B Model Match 1T-Parameter Rivals

Apodex has released an open-source agent stack aimed at closing the gap between chatbots that write summaries and systems that finish work. The release bundles FrontierAgent, an Apache 2.0 terminal agent framework, alongside Apodex 1.1 mini, a 35B open-weight model tuned to drive it. Install one command, point it at a local endpoint or the hosted API, and hand it long-horizon research or file-based tasks that produce verifiable deliverables.

Two agents share a terminal

FrontierAgent is a TUI (terminal user interface) that ships two workflows. ReAct runs one stateful agent that researches, reads files, writes deliverables, runs commands, and iterates inside a task-scoped sandbox. Agent Team runs a coordinator that maintains a task board, delegates independent work to parallel sub-agents, collects their reports, and synthesizes the result. The same engine powers the benchmark harness, so the runtime you use locally is the one used to score the models.

Agent Team is the more interesting design. The coordinator decomposes the request, dispatches bounded parallel assignments, receives structured reports, and can call an optional fast reporter for final evidence review. Task board updates appear live in the TUI sidebar with pending, active, completed, blocked, and cancelled states, so you can watch what each sub-agent is doing without tailing a log file.

Steer mid-run without a restart

Asynchronous intervention deserves a callout for anyone who has fought with agent frameworks. Typing while an agent is running queues a new instruction that gets injected at the next safe turn boundary without discarding the active run. In Agent Team mode it steers the coordinator, while already-running sub-agents finish their assignments. You can drop a new requirement halfway through a research task and the coordinator replans only what is affected instead of starting over.