Guides
The long answers.
Longer than an FAQ answer and shorter than the paper. Each guide opens with the short answer, states its numbers with the qualifier they are only true under, and names where each figure came from.
Start with the symptom
What the agent is doing right now, why it does it, and what stops it. Written for someone reading this with a runaway process open in another window.
- 06
Why an agent gets stuck in a loop
Diagnose an agent that retries forever: what a saturating loop looks like, how to confirm it from your logs, what it costs while you debug, and how to stop it.
7 sections - 07
Why an agent burns tokens
Work out where an LLM agent bill went: why cost grows faster than turns, how to read the spend, which patterns cause it, and what actually reduces it.
7 sections
Then the mechanism
How runtime control works, what it is measured against, and where it stops. Written for someone deciding whether the approach holds before installing anything.
- 01
Deterministic control for LLM agents
What deterministic control means for a language model, why the control layer belongs outside the model, how determinism is verified, and what to install today.
9 sections - 02
Preventing agent loops and runaway spend
Why LLM agents loop, what a runaway loop costs across seven measured configurations, how to detect one, and where a spend cap stops being enough.
10 sections - 03
Jailbreaking and prompt injection
Why prompt injection works, what a token-free control boundary is, exactly what it does and does not protect, and how to test the boundary yourself.
7 sections
- 04
Auditing and testing LLM agents
What makes an LLM agent auditable, the five things an audit should ask for, how to replay and re-score a published record, and what to be sceptical of.
8 sections - 05
A governor compared with prompt engineering
How a deterministic governor differs from prompt engineering, which one contained a runaway loop in a pre-registered ablation, and where prompting still wins.
7 sections