Inside Apodex 1.1: Adaptive Recovery, AgentOS, and self-organizing async agent teams

testingcatalog · x · 2026-09-16

TestingCatalog's deep dive on Apodex 1.1. Instead of reading web pages and returning prose, the new version opens files directly — spreadsheets, PDFs, datasets — picks a method, writes and runs the analysis code, and returns tables, charts, and scripts traceable to source. It went live August 24.

The core problem it targets is the cost of long tasks collapsing midway: an agent that hits a missing value or failed assumption at step forty either stops or returns something vague, wasting hours already spent. Apodex's answer is Adaptive Recovery, where the system explains the adjustment and continues. Two training directions sit behind it: Environment Scaling widens the file/search/code environments the model learns in, while Agentic Coordination Scaling teaches task decomposition. Both run on AgentOS, a runtime holding tool calls, file state, and whole-job progress.

The headline Deep Discover mode lets the model assemble an Agent Team on its own — deciding whether to split, how many subagents to run, and when to consolidate. The team is asynchronous: branches report intermediate results to shared task state as they finish, and mid-run file additions or requirement changes trigger replanning of only the affected parts.

Original post →

Inside Apodex 1.1: Adaptive Recovery, AgentOS, and self-organizing async agent teams

More from coding & agent