Skip to content

Agent & Developer Resources

WhiteBoxXAI is used by people and by the AI agents working for them. This section collects the entry points an agent needs to set up WhiteBoxXAI, call it correctly, and report what it returned without inventing anything.

Start here (for agents)

Resource URL Use it to
Docs index for LLMs https://docs.whiteboxxai.com/llms.txt Find the right docs page without crawling the whole site
Agent skills index https://whiteboxxai.com/skills/index.md Pick which skill to load for a task
Skills discovery (JSON) https://whiteboxxai.com/.well-known/skills/index.json Auto-discover and verify skills in a harness
All skills (zip) https://whiteboxxai.com/skills/whitebox-skills.zip Install every skill at once
MCP server MCP Server Give an assistant live tools over the user's WhiteBoxXAI data
Compliance-assistant prompt AI Prompts Bootstrap an assistant that summarises evidence for a risk officer
Client config files Agent Config Definitions Copy-paste MCP client configuration

The fastest way to get a coding agent going is to paste this instruction into it:

Read https://whiteboxxai.com/.well-known/skills/whitebox-getting-started/SKILL.md and set up WhiteBoxXAI for this repository.

Which integration path?

You are... Use Docs
Instrumenting Python code that trains or serves a model Python SDK SDK guide
Instrumenting an LLM app, RAG pipeline or agent Python SDK LLM monitoring, Multi-agent monitoring
An assistant that should query or act on WhiteBoxXAI for a user MCP server MCP Server
Working in a language other than Python REST API under /api/v1/ API Reference

Rules for agents

These are the same rules the whitebox-governance-rules skill enforces. They exist because WhiteBoxXAI output becomes compliance evidence that a person may hand to an auditor.

  1. Never invent a number. If a call returns nothing, an error, or an unassessed state, say exactly that. A wrong score presented as a result is worse than no score.
  2. Put a plain-language verdict next to every ML number, using the platform's own verdict when it returns one. Don't present your own pass/fail threshold as the platform's.
  3. Keep secrets out of code. Read the key from the environment (WHITEBOX_API_KEY for the SDK, WHITEBOXXAI_MCP_API_KEY for the MCP server). Keys look like wbx_live_....
  4. A 404 means "not in your organization". Don't retry with other credentials or guess ids.
  5. Human decisions stay human. Governance votes need a person's own session, two-factor authentication and an authenticator code. An organization API key is refused. Prepare and summarise the dossier; never vote on someone's behalf.
  6. Stay inside the five frameworks: ISO/IEC 42001, GDPR, CCPA, the EU AI Act and NIST AI RMF. WhiteBoxXAI records evidence; it does not certify compliance.
  7. Meter responsibly. Group one AI run's records with governance_run() and honour Retry-After on a 429. See Governance Events.