Documentation

Overview

What PrysmAI is, the two integration surfaces it exposes, and the shared control plane behind them.

Getting Started

The shortest developer path: create a project, connect a provider, generate a Prysm key, and send your first request.

Python SDK

The current Python golden path: connect through the proxy, attach metadata, or use MCP-backed session helpers.

Governance

How governance now fits the product: one session/evidence model across MCP and proxy-backed activity.

Framework Integrations

LangGraph, CrewAI, and LlamaIndex integrations with optional governance.

CI/CD Integration

Dynamic upstream API keys, custom header forwarding, and CI/CD pipeline integration with GitHub Actions and GitLab.

Dashboard

Overview metrics, request explorer, API keys page, and settings.

API Keys

Create, view, revoke, and rotate API keys for proxy authentication.

Alerts

Metric-based alerting with email, Slack, Discord, PagerDuty, and webhook channels.

Team Management

Invite members, assign roles, and manage organization access.

Cost Tracking

Automatic cost calculation, custom model pricing, and usage limits.

Security

Prompt injection detection, PII redaction, content policies, and threat scoring.

Explainability

Token confidence heatmaps, hallucination detection, and decision point analysis.

REST API

OpenAI-compatible proxy endpoints for chat completions, completions, and embeddings.

Providers

Configuration for OpenAI, Anthropic, Google Gemini, and custom/self-hosted models.

Self-Hosted Proxy

Running the Prysm proxy on your own infrastructure.

All Endpoints

Complete reference of all proxy endpoints including WebSocket live feed.

Response Headers

Security scan headers and rate limit headers returned on every proxy response.

Advanced Features

Tool calling, logprobs, and the complete list of captured data fields.

Example Applications

Full working applications demonstrating Prysm AI capabilities.

Error Handling

Error types, Prysm-specific errors, and common troubleshooting patterns.

Changelog

Version history and release notes for the Prysm AI SDK and platform.

Microsoft Agent Framework

Integrate Prysm with Microsoft Agent Framework for full agent observability.

Unified Timeline

Correlated view of LLM traces, tool events, and session events.

Tool Performance

Monitor tool execution metrics, success rates, and latency patterns.

Agent Decisions

Understand why your agent chose specific tools and actions.

Workflow Graph

Directed execution graph showing node-to-node flow for agent sessions.

Financial Anomaly Detection

Detect when an agent's cost trajectory deviates from baseline and alert or halt execution.

Resource Access Detection

Define an agent's allowed resource envelope and detect out-of-scope access in real time.

Agent Loop Detection

Detect circular tool call patterns and repetitive state transitions with configurable circuit breakers.

Multi-Agent Coordination Monitoring

Track inter-agent communication, detect conflicting instructions, and view agent network topology.