Waxell
Waxell is an AI agent governance and observability platform that provides runtime policy enforcement, auto-instrumented LLM telemetry, MCP governance, cost management, and durable workflow execution for agents built in any Python framework or third-party agentic tool (Claude Code, Cursor, LangChain, CrewAI, OpenAI Agents SDK, and 200+ more).
APIs
Waxell Observe API
The Waxell Observe REST API exposes the AI agent governance and observability control plane. It is used by the waxell-observe Python SDK and the Developer MCP server to record r...
Waxell Developer MCP Server
Waxell Developer MCP is a hosted Model Context Protocol server that lets coding agents (Claude Code, Cursor, Windsurf, VS Code, Claude Desktop) query a Waxell instance in real t...
Collections
Waxell Observe API
OPENPricing Plans
Rate Limits
FinOps
Waxell Finops
FINOPSFeatures
Two-line setup auto-instruments 200+ libraries (OpenAI, Anthropic, LangChain, LlamaIndex, CrewAI, LiteLLM, etc.) without code changes.
26 policy categories (cost, kill switch, PII, compliance, scope, safety) returning seven decisions (allow, warn, redact, throttle, block, skip, retry).
Auto-instrumentor, server middleware, and governance proxy for Model Context Protocol traffic with PII scanning and rug-pull detection.
Built-in model pricing for 20+ models, tenant overrides via REST, budget enforcement that warns/throttles/blocks at thresholds.
Versioned managed prompts retrievable by name and label (e.g. production, staging) directly from the SDK.
Durable execution boundary with checkpoint and resume; Redis-backed in production, in-memory for development.
Custom handlers route policy blocks to Slack, webhooks, or terminal prompts for human review.
Immutable, timestamped record of all agent actions, decisions, and governance events.
Hosted SSE MCP server (dev-mcp.waxell.dev/sse) with 15 live tools and 8 docs resources for coding agents.
PII fields encrypted at the application layer with AES-256-GCM and AWS KMS (FIPS 140-2 Level 3) before database storage.
Use Cases
Enforce policies on Claude Code, Cursor, Windsurf, VS Code, and Claude Desktop without modifying their code.
Add full observability to LangChain, CrewAI, OpenAI Agents SDK, or custom Python agents with the @waxell.observe decorator.
Set budgets on token spend per agent, user, or tenant; block runs that exceed configured limits.
Scan MCP tool inputs/outputs for PII, credentials, and secrets with warn/block/redact responses.
Use the WorkflowEnvelope to checkpoint multi-step agent workflows so they can resume after interruption.
Maintain SOC 2 Ready posture with immutable audit trails, encrypted PII, and EU data residency.
Integrations
Auto-instrumented LLM provider; cost and token tracking out of the box.
Auto-instrumented LLM provider; supports Claude family models.
First-class callback handler (WaxellLangChainHandler) for tracing chains and graphs.
Auto-instrumented multi-agent framework support.
Tracing for RAG pipelines built with LlamaIndex.
Unified telemetry across LiteLLM-routed providers.
Governance overlay for Anthropic's Claude Code coding agent via the Developer MCP.
Coding-agent governance via the SSE MCP server at dev-mcp.waxell.dev.
Auto-instrumentation for OpenAI's Agents SDK runs.
Cloud LLM providers covered by Waxell's auto-instrumentation.
Vector database call tracing with retrieval span recording.
Human-in-the-loop approval handlers for policy blocks.
Listed subprocessor for billing.
Solutions
Govern third-party agents (Claude Code, Cursor) without code changes via the MCP governance proxy.
Instrument self-built agents with auto-instrumentation, policy enforcement, and cost attribution.
Governed execution environment for high-risk workflows with the durable WorkflowEnvelope.