Hollow

v4.1.1 Breaking

This release includes 2 breaking changes for platform teams planning a safe upgrade.

Published 2mo AI Agents & Assistants

✓ No known CVEs patched

✓ No known CVEs patched in this version

Summary

AI summary

Default model switched to qwen3.5:9b non-thinking mode, enforcing think=false on all Ollama endpoints.

Full changelog

qwen3.5:9b non-thinking mode — switched default model, think=false enforced on all Ollama endpoints so agents never stall on empty thinking responses
Agent cap — daemon now runs max 12 concurrent agents instead of unbounded, eliminating Ollama contention
Goal diversity — follow-on goal prompt now includes last 10 completed goals with explicit "do not repeat" instruction
Idle agents get new work — when an agent finishes with no follow-on goal, the daemon immediately assigns one of 7 rotating exploratory prompts (filtered against the agent's recent history)
TUI agent count fixed — agent list rebuilds fresh each cycle, completed agents drop off instead of accumulating
Orchestrator agent — dedicated root agent running priority-10 goal to review completed work and propose 5 new diverse research directions

Default Ollama model changed to qwen3.5:9b with non-thinking mode enforced (think=false) on all endpoints.
Maximum concurrent agents limited to 12, removing the previous unbounded behavior.

Weekly OSS security release digest.

The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.

No spam, unsubscribe anytime.

Share this release

Track Hollow

Get notified when new releases ship.

About Hollow