ramen-ai Integrations

The deterministic execution boundary for AI agents.

---

The Problem

LLMs cannot police their own tools. If your agent is connected to a database, a prompt injection hidden in a PDF can drop your tables. If it is connected to a payment API, a Morse-encoded instruction buried in a retrieved document can drain a wallet. Standard keyword filters catch basic syntax. They fail against encoded payloads, corporate jargon, and multi-criterion composite attacks.

We built the deterministic execution boundary to stop it — a semantic firewall that evaluates every tool call against your compliance policies before execution, signs every verdict with an Ed25519 receipt, and returns a steering instruction that is auditable, reproducible, and legally defensible.

---

               

---

Can you bypass it?

Standard safety filters catch basic syntax. They fail against encoded payloads and corporate jargon. We challenge you to bypass our semantic firewall using the zero-day evasion vectors in our official Red Team Guide.

Below is a simulation of the Grok/Bankr heist. We fed the raw adversarial prompt directly into our sandbox. It uses a social engineering wrapper (claiming a visual impairment) to smuggle a 3,000,000,000 DRB transfer instruction encoded in Morse code. The firewall evaluated the underlying semantic intent, intercepted the unauthorized financial transfer, and blocked it pre-execution, issuing a verified Ed25519 receipt.

---

Getting Started

To use these integrations, you must mint an API Key.

We offer a Free Starter Tier (1,000 evaluations/month, BYOK) which includes full access to our Core IT Security bundle. Mint your key at:

https://ramenai.dev/pricing

BYOK — Bring Your Own Key

The Starter and Professional tiers use your own LLM provider key for inference. You need two keys:

Key	Purpose	Where to get it
RAMEN_API_KEY	Authenticates you to the ramen-ai platform	ramenai.dev/pricing
OPENAI_API_KEY (or Anthropic equivalent)	Forwarded as X-Provider-Key for LLM inference	Your provider's developer portal

export RAMEN_API_KEY=ramen_ak_...
export OPENAI_API_KEY=sk-...        # or ANTHROPIC_API_KEY, etc.

Enterprise tier users have keys managed server-side — omit providerKey entirely.

---

Integrations

Core SDKs

Use these as the foundation for custom integrations across any framework, queue, or gateway not covered by the plugins below.

Package	Language	Path	Description
@ramen-ai/node-core	TypeScript / Node.js ≥ 18	core-clients/node/	Zero-dependency HTTP client using Web Crypto. Covers RamenClient, BYOK, standalone verifyReceipt, and injectable fetchImpl for testing. → Full SDK docs
ramen-ai-core	Python ≥ 3.10	core-clients/python/	Synchronous httpx client with cryptography Ed25519 verification. Per-call BYOK, standalone verify_receipt, and pytest-httpx test patterns. → Full SDK docs

Plugins

Plugin	Platform	Path	Description
agt-typescript	Microsoft AGT	plugins/agt-typescript/	TypeScript middleware wired as an AGT ExternalPolicyBackend. Intercepts tool calls pre-execution, verifies receipts, logs to the AGT audit chain.
github-action	GitHub Actions	plugins/github-action/	Scans PR diffs for system-prompt modifications, evaluates against compliance policies, fails CI on [BLOCKED] — with a cryptographic receipt comment on the PR.
langchain-python	LangChain (Python)	plugins/langchain-python/	BaseCallbackHandler that intercepts LangChain tool calls pre-execution and halts the chain on [BLOCKED].
pydantic-ai	PydanticAI (Python)	plugins/pydantic-ai/	args_validator factory that intercepts tool calls after schema validation and halts the agent run on [BLOCKED].
mcp-proxy	MCP stdio (universal)	plugins/mcp-proxy/	Universal stdio proxy intercepting tools/call JSON-RPC at the transport layer. Works with Claude Desktop and any stdio MCP client.
cmcp-python	cMCP + TRACE	plugins/cmcp-python/	cMCP tool-call policy adapter, plus a TRACE Trust Record exporter that maps V5 Ed25519 receipts onto the agentrust-io EAT profile.
mlflow-python	MLflow / Databricks	plugins/mlflow-python/	mlflow.pyfunc.PythonModel wrapper enforcing algorithmic governance on classical ML. Evaluates feature arrays and SHAP attributions for proxy bias pre-inference; halts serving on [BLOCKED].
ramen-data-filter	Pandas / CSV	plugins/ramen-data-filter/	Dual-mode row filtration for RAG ingestion and MLOps datasets. Strictly excludes blocked records or semantically imputes steering-approved columns.
dsh-ramen-guard	DeepSeek Harness	plugins/dsh-ramen-guard/	Unofficial Cordis guard that, in enforcement mode, blocks policy-violating tool intent before execution and requires a locally verified Ed25519 receipt for allowed calls.

---

How it works

Every evaluation returns a V5 Ed25519 cryptographic receipt — a signed, self-describing record binding the verdict to a SHA-256 hash of your input. Receipts are verified locally against the published public key: no trust in the API server is required.

Your agent  →  ramen-ai middleware  →  POST /api/v1/paas/evaluate
                                              ↓
                                     Verdict + Ed25519 receipt
                                              ↓
                             Local receipt verification (Web Crypto)
                                              ↓
                              ALLOWED → tool executes
                              BLOCKED → thrown / build failed / PR comment

---

Bundles

Use bundle_ids for macro-level Defence-in-Depth aggregation — a single slug activates multiple coordinated policies across a threat domain. For surgical, single-domain statutory auditing, pass specific policy_ids directly to route to exactly one policy without other classifiers interfering.

The Free Starter Tier includes the Core IT Security bundle:

Bundle slug	Coverage
ramen__shield_core_it	Destructive execution, infrastructure abuse, prompt leakage & jailbreak, secret exfiltration, OWASP ASI-06 indirect injection
ramen__eu_ai_act_baseline	EU AI Act Articles 5, 10, and 50 — prohibited practices, data governance, transparency obligations

Full bundle and policy reference: https://ramenai.dev/pricing

---

Repository structure

/
├── assets/                  # Shared brand and visual assets
├── core-clients/
│   ├── node/                # @ramen-ai/node-core  — TypeScript SDK
│   └── python/              # ramen-ai-core        — Python SDK
├── plugins/
│   ├── agt-typescript/      # Microsoft AGT middleware
│   ├── github-action/       # GitHub Actions CI/CD interceptor
│   ├── langchain-python/    # LangChain Python callback handler
│   ├── pydantic-ai/         # PydanticAI args_validator middleware
│   ├── mcp-proxy/           # Universal MCP stdio transport interceptor
│   ├── cmcp-python/         # cMCP policy adapter + TRACE record exporter
│   ├── mlflow-python/       # MLflow pyfunc algorithmic governance wrapper
│   ├── ramen-data-filter/   # Pandas/CSV filtration for RAG and MLOps datasets
│   └── dsh-ramen-guard/    # DeepSeek Harness pre-execution Cordis guard
├── .github/
│   └── workflows/           # CI workflows
└── RED_TEAM_GUIDE.md        # Zero-day evasion vectors and challenge guide

---

Adding integration logos

When adding a platform integration, update the standard ecosystem row in the home README, every plugin README, and every translated companion. See Adding Integration Logos for branding, internal-link, self-badge, asset, disclosure, and validation requirements.