# PURISTA AI Harness

A production-oriented harness for typed models, tools, agents, memory, evaluation, and observability.

---
Canonical: /harness/
Source: web/src/data/harness-markdown.ts
Format: Markdown for agents
---

PURISTA AI Harness organizes AI capability as explicit application architecture instead of ad hoc prompt calls.

Use it when an application needs typed tool boundaries, model capability routing, memory control, reproducible tests, and observable agent behavior.

## Core Ideas

- Define models by capability, not by scattered provider calls.
- Expose tools with typed input and output contracts.
- Build agents as named, testable runtime components.
- Keep prompts, tools, memory, and policy boundaries reviewable.
- Treat security, privacy, and evaluation as part of the architecture.

## Recommended Path

1. Start with the harness mental model.
2. Define provider models and capability names.
3. Add narrow tools with explicit schemas.
4. Compose agents around business use cases.
5. Add tests and evaluations before production rollout.
6. Connect observability and review failure paths.