perathos

integrations

On both sides of your AI stack.

Perathos verifies responses from every major LLM provider. It also installs as an MCP server so agent CLIs and IDE assistants can invoke verification on demand. Same proof bundle format. Same verdict semantics. Different entry points.

upstream

Models we verify

Any LLM with an OpenAI-compatible endpoint. Drop in a single base_url swap and every response is wrapped in a VRL Proof Bundle.

Anthropic

Claude family

available

Claude Opus, Sonnet, and Haiku — including the latest 4.x generation. Verified via base_url swap on the official Anthropic SDK or any OpenAI-compatible client.

OpenAI

GPT + o-series + Codex

available

GPT-series chat models, o-series reasoning models, and Codex coding model. Native OpenAI SDK is supported by changing the base_url; no SDK change required.

Google

Gemini · Vertex AI

available

Gemini Pro and Flash via the OpenAI-compatibility endpoint or via Vertex AI. Multi-modal claims (image, structured JSON) supported by the Schema Validator.

Meta

Llama family

available

Llama 3 and Llama 4 via any hosted endpoint (AWS Bedrock, Groq, Together, Fireworks) or self-hosted vLLM / TGI deployments.

OpenRouter

100+ open-source models

available

Mistral, Qwen, DeepSeek, Nemotron, Yi, and dozens more. One integration covers every model in the OpenRouter catalog.

Self-hosted

vLLM · TGI · Ollama · custom

available

Any endpoint that speaks the OpenAI chat-completions protocol. Including fine-tuned and proprietary models. The Model Fingerprinter binds responses to your specific deployment.

Azure OpenAI

Enterprise Azure deployments

available

Native support for Azure OpenAI deployments with regional endpoints, private networking, and Azure AD authentication.

AWS Bedrock

Multi-model AWS

available

Claude, Llama, Mistral, and Titan accessed through Bedrock. AWS IAM-authenticated requests verified inline.

Groq

Low-latency inference

available

Llama 3.x and Mixtral on Groq's LPU hardware. Verified at full pipeline or fast-path latency depending on workload.

downstream

Clients that invoke us

Perathos ships an MCP server so any agent CLI or IDE assistant can request a verification verdict the same way it calls any other tool.

Claude Code

Anthropic CLI · MCP

available

The official Anthropic CLI for coding workflows. Perathos installs as a one-line MCP server, then Claude can verify any response, citation, or calculation it produces with a tool call.

claude mcp add perathos -- npx -y @perathos/mcp-server

OpenAI Codex CLI

OpenAI CLI · tool integration

available

OpenAI's coding agent CLI. Perathos registers as a configured tool — Codex can verify generated code against the Symbolic Solver and Schema Validator before applying changes.

codex tool add perathos --url https://api.perathos.com/v1/verify

Cursor

IDE · MCP

available

Verification inline in the editor. Cursor can attach a verdict to any AI suggestion before insertion. Same MCP server install as Claude Code.

# In Cursor: Settings → MCP → Add Server
{ "perathos": { "command": "npx", "args": ["-y", "@perathos/mcp-server"] } }

Continue

VS Code / JetBrains · MCP

available

Open-source AI coding assistant. Adds Perathos as a verifier so generated code is checked before being applied to the file.

# In ~/.continue/config.json
{ "mcpServers": { "perathos": { "command": "npx", "args": ["-y", "@perathos/mcp-server"] } } }

Cline / Roo / Aider

Agentic CLIs · MCP

beta

Other popular agent CLIs that speak MCP. Same install pattern — bring your own client config. Open an issue if your tool needs a specific transport.

LangChain · LangGraph

Framework · Python / TypeScript

available

Drop-in Python and TypeScript modules. Verification becomes a node in your graph; verdicts gate downstream tool calls or trigger human review.

pip install perathos
from perathos import verify
verdict = verify(response)

LlamaIndex

Framework · RAG pipelines

available

Post-retrieval verification for RAG outputs. Confirms cited sources exist and supports the claim before the response reaches the user.

Your custom agent

Any framework · REST + SDK

available

Direct REST API and native Python / Node SDKs. If your agent can make an HTTP call, it can call Perathos. Streaming verification supported.

open standards

Standards we implement

Model Context Protocol

Anthropic's open MCP spec — the standard for AI agents to discover and call external tools. Perathos publishes a compliant server.

OpenAI chat-completions

The de-facto protocol for LLM inference. Any client speaking this protocol works with Perathos via a one-line base_url change.

VRL Proof Bundle

Our open output format. Apache-2.0 licensed schema, conformance vectors, and reference verifier. Built so reviewers can audit verdicts without trusting us.

Don't see your stack?

Anything OpenAI-compatible is supported out of the box. For custom or proprietary protocols, we'll build the adapter during your pilot.

Request integration →Read the docs