Build with Pointsman

Pointsman is an open-source gateway between your applications and model providers. Use it to manage virtual keys, model access, spending controls, caching, guardrails and routing through a documented Chat Completions API.

Get started#

A local walkthrough is an evaluation setup. Read deployment before exposing a gateway. This branch contains a performance candidate awaiting long-run requalification. The earlier qualified release scope and retained limitations are described in release status.

Understand the design#

Start with the architecture and request lifecycle, which connects the system diagram to the order of a real request. Explore each part of the design:

• Authentication and identity
• Policy-based authorization: scopes, grants and inheritance
• Rate limiting, budgets and settlement
• Guardrails
• Exact and semantic caching
• Routing and adaptive selection
• Policy service and durable learning

Each guide explains why the design exists, its state and trust boundaries, failure behavior, operational checks and tradeoffs. Adaptive and quality-based paths remain experimental and opt-in; weighted-random is the default.

Configure your gateway#

Connect providers and models, manage provider credentials, issue virtual keys and budgets, or build a team hierarchy.

Use traffic weights, guardrail rules and cache tuning to match your application's needs. Learn where configuration lives before changing it.

Reference and operations#

• Application API, admin API and response headers.
• Configuration keys and operational limits.
• Request troubleshooting, operating notes, security and known issues.
• Latency and benchmark breakdown.
• Release scope and qualification.

The GitHub repository contains the project source. These public guides focus on using, understanding and operating the product.