Scalarr

Parallax Evaluation API

Self-hosted HTTP API of Parallax, the Apache-2.0 runtime security engine Scalarr publishes for AI agents. Two endpoints: POST /evaluate takes a lifecycle event (message.before, tool.before, tool.after, params.before) with optional tool_name, tool_args, tool_result, message_text, model and params, runs it through a cost-ordered chain of regex, keyword-pattern, Sigma, CEL and SQL evaluators, and returns a decision (block, redact, detect or allow) with the reasons that matched and the elapsed evaluation time; GET /health returns status, mode, evaluator count and version. A proxy mode fronts the Anthropic Messages API and evaluates traffic transparently. The server ships with no authentication and binds to 127.0.0.1:9920 by default.

GitHub

Documentation

📖
Documentation
https://github.com/agent-defense/parallax/tree/main/docs
📖
APIReference
https://github.com/agent-defense/parallax/blob/main/docs/integrations/generic.md
📖
GettingStarted
https://github.com/agent-defense/parallax#-quick-start

Other Resources

🔗
SourceCode
https://github.com/agent-defense/parallax
🔗
Roadmap
https://github.com/agent-defense/parallax#-roadmap
🔗
Webhooks
https://raw.githubusercontent.com/api-evangelist/scalarr/refs/heads/main/asyncapi/scalarr-parallax-webhooks.yml
🔗
CLI
https://raw.githubusercontent.com/api-evangelist/scalarr/refs/heads/main/cli/scalarr-cli.yml
🔗
APIsJSON
https://raw.githubusercontent.com/api-evangelist/scalarr/refs/heads/main/apis.yml

API entry from apis.yml

apis.yml Raw ↑ 
name: Parallax Evaluation API
description: 'Self-hosted HTTP API of Parallax, the Apache-2.0 runtime security engine Scalarr publishes
  for AI agents. Two endpoints: POST /evaluate takes a lifecycle event (message.before, tool.before, tool.after,
  params.before) with optional tool_name, tool_args, tool_result, message_text, model and params, runs
  it through a cost-ordered chain of regex, keyword-pattern, Sigma, CEL and SQL evaluators, and returns
  a decision (block, redact, detect or allow) with the reasons that matched and the elapsed evaluation
  time; GET /health returns status, mode, evaluator count and version. A proxy mode fronts the Anthropic
  Messages API and evaluates traffic transparently. The server ships with no authentication and binds
  to 127.0.0.1:9920 by default.'
humanURL: https://github.com/agent-defense/parallax
baseURL: http://{host}:{port}
tags:
- Runtime Security
- AI Agent Security
- Open Source
- Prompt Injection
- Policy Enforcement
properties:
- type: Documentation
  url: https://github.com/agent-defense/parallax/tree/main/docs
- type: APIReference
  url: https://github.com/agent-defense/parallax/blob/main/docs/integrations/generic.md
- type: GettingStarted
  url: https://github.com/agent-defense/parallax#-quick-start
- type: SourceCode
  url: https://github.com/agent-defense/parallax
- type: Roadmap
  url: https://github.com/agent-defense/parallax#-roadmap
- type: Webhooks
  url: asyncapi/scalarr-parallax-webhooks.yml
- type: CLI
  url: cli/scalarr-cli.yml