Nexuspoint Tools

Context management that survives an audit.

0.19.3 TypeScript · MIT for the SDK, commercial for the hosted control plane

What it does

Content boundaries

Retrieved content is tagged as untrusted at the point of retrieval and stays tagged through the pipeline, so a policy can act on it.

Deterministic replay

Reconstruct a historical run with the exact context it saw. Useful for debugging, necessary for incident review.

Redaction hooks

Run a transform over content before it enters the window, with the original and transformed form both recorded.

Token accounting

Per-run breakdown of which source consumed which share of the window, which is usually where the cost went.

Install

Node
npm install @agent/context --registry https://registry.ai-devtools.eu
Python
pip install agent-context --index-url https://pypi.ai-devtools.eu/simple
Replay CLI
curl -fsSL https://ai-devtools.eu/install.sh | sh

Background

We are deliberately not a model provider and not an agent framework. We sit alongside whichever you have chosen and record what crosses the boundary. Teams switch frameworks more often than they expect, and the audit trail should survive that.

Read the documentation