What is the Blunom Agentic Router?

It is an AI gateway with governance built in. It gives your applications one endpoint to call any approved model or tool, while Blunom enforces your security and spend policy on every request.

Is this an AI gateway?

Yes, and it is an AI gateway you deploy and manage: one control point for all of your AI, covering model and tool access, governance, and cost controls. You keep your own provider keys and policies, and every model call and tool request is enforced against your rules.

Do I have to rewrite my code to use it?

No. It speaks the OpenAI, Anthropic, and AWS Bedrock SDK interfaces. You change the base URL and use a Blunom key; your existing request and streaming code works unchanged.

What API endpoints does the Agentic Router expose?

One key and one base URL expose the full set of primitives: inference, agents, tools, knowledge, governance, and events. The One API. Every primitive. section on this page lists each one, from Chat Completions and agent runs to guardrails, usage, batches, and webhooks.

Can I run agents, tools, and embeddings through the same key?

Yes. Agents, tools, embeddings, and the rest of the primitives use the same key and the same governance: your AI firewall, your policies, spend caps, and one audit row with a trace id.

Can I switch or add models without changing my application?

Yes. Models are selected by an allow-list and an optional fallback chain, so you can adopt a newer or cheaper model, or add automatic failover, without an app change.

Does it use my own provider accounts?

Yes. You bring your own provider keys, so you keep your negotiated pricing. Credentials are resolved server-side and never exposed to your client.

How does it keep AI usage secure and compliant?

Every request passes a built-in AI firewall (PII, credit-card, and SSN redaction, plus prompt-injection detection) and your content policy, and writes one audit entry with a trace id. You also set per-key spend caps and model and action allow-lists.

Can it call tools and business applications, not just models?

Yes. A single governed key can run approved actions across your connected integrations through the same policy and audit trail as model calls.

Part of Blunom OS

Blunom Agentic RouterBuilt-in AI Firewall and Real-time TokenOps

Your own AI gateway that routes every model and tool, with governance built in, plus an AI Firewall and real-time TokenOps

Book a demo

EXISTING SDKs
SDKs: OpenAI, Anthropic, and Bedrock
AI FIREWALL
Your policies, enforced on every call and tool request
TOKENOPS
Full audit trail, trace id on every request
ANY MODEL
350+ models across every major provider

Our Customers and Partners

AI IS FRAGMENTED

Blunom can help with model routing, governance, tool access controls, and token efficiency.

Book a demo

Compatible with what you already run

Models and Providers

Route across every major provider: OpenAI, Anthropic, Google Gemini, AWS Bedrock, xAI (Grok), DeepSeek, Mistral, and Perplexity. Bring your own trained and open-weight models, too.

Tools

Run governed actions across 100+ integrations in the Blunom Tool Catalog, including Google Workspace, Slack, Salesforce, Snowflake, and more.

SDKs

Use the official OpenAI SDK, the Anthropic SDK, or the AWS Bedrock SDK (boto3). An OpenAPI specification is published for typed client generation.

Point your SDK at Blunom. Keep your code.

The Agentic Router sits between your application and the model providers. You change a base URL and use a Blunom key. Everything else, the request shape, the streaming, the tool-call round-trip, stays exactly as your SDK already does it.

Connect once

Update your existing SDK code to point to Blunom, and bring your own provider keys (BYOK) as integration profiles. Keep your negotiated provider discounts.

Route anything

Use any model from any provider. Your admin chooses which models are enabled, and Blunom handles the credentials, routes each request, and falls back automatically if a provider is unavailable.

Govern everything

Every call passes your AI firewall and content policy, with real-time TokenOps spend governance and observability, and writes one audit row with a trace id.

Before: a key and an SDK for every provider

from openai import OpenAI
from anthropic import Anthropic

openai = OpenAI(api_key="sk-...")          # provider keys live in your app
claude = Anthropic(api_key="sk-ant-...")
# plus Bedrock, Gemini, more keys
# no firewall, no spend caps, no audit trail

After: one Blunom key, any model, every call governed

from openai import OpenAI

client = OpenAI(base_url="https://app.blunom.ai/api/v1",
                api_key="blu_live_...")    # provider keys stay in Blunom

resp = client.chat.completions.create(
    model="claude-sonnet-4-5",             # Claude, through the OpenAI SDK
    messages=[{"role": "user", "content": "Draft a Q3 update for jane@acme.com"}],
    extra_body={"blunom": {"models": ["claude-sonnet-4-5", "gpt-5.5"]}},  # fallback
)

Works with the OpenAI SDK, the Anthropic SDK, and the AWS SDK (Boto3), each with its own Blunom base URL.

Everything a control plane for AI should do

Blunom Agentic Router is a drop-in API that speaks the OpenAI, Anthropic, and AWS Bedrock SDKs. Route across every major model and provider, run governed tool calls across your business apps, and enforce one AI firewall and spend policy on every request. No refactoring.

Route Any Model

One endpoint for every provider, with fallback and a hot-swap when a better model shows up.

Govern Every Call

One firewall, one spend policy, and one audit trail on every request.

Run Tools and Knowledge

The same key reaches approved business apps and knowledge bases.

One API. Every primitive.

One Blunom key and one base URL unlock a full, SDK-compatible API. Every primitive is governed the same way: your AI firewall, your policies, spend caps, and one audit row with a trace id on every call. Add a capability without adding a new trust boundary.

Inference, in the SDKs you already use

Chat Completions/v1/chat/completions
OpenAI-compatible chat with streaming and tool calls. Route any chat model without changing your code.
Messages/v1/messages
Anthropic-compatible. Point the Anthropic SDK at Blunom and keep the same governance.
Bedrock/v1/bedrock/...
AWS Bedrock-compatible invoke. Keep boto3, gain the firewall and audit.
Embeddings/v1/embeddings
OpenAI-compatible vectors through one governed endpoint.
Rerank/v1/rerank
Cohere-compatible relevance reranking to sharpen retrieval results.
Models/v1/models
List the exact models this key is allowed to call.

Agents, not just models

Invoke an agent/v1/agents/{id}/invoke
Run a full governed agent (planning, tools, knowledge) and stream the result, like calling a model but with real work done.
Async runs/v1/agents/{id}/runs
Kick off long-running work, then poll /v1/runs/{id} or stream /v1/runs/{id}/events, and cancel anytime. Durable background jobs with live progress.
Schedules/v1/agents/{id}/schedules
Run an agent on a recurring cron cadence, under the same governance as an interactive call.
Discover agents/v1/agents
List the agents a key can invoke, no copying IDs from a dashboard.

Tools and actions

Tool catalog/v1/integrations
List the governed integration actions this key can run.
Run an action/v1/integrations/{app}/{action}
Call one approved action on a connected app (send a Slack message, run a Snowflake query) with credentials kept server-side and the same policy as a model call. Deterministic, no model required.

Knowledge and memory

Knowledge search/v1/knowledge-bases/{id}/search
Semantic search inside an approved knowledge base, access-scoped and opt-in.
Documents/v1/knowledge-bases/{id}/documents
Add and manage the content an agent can ground on.
Memory/v1/knowledge-bases/memory
Inspect or clear what a key has learned, so you control what persists.

Governance as a primitive

Guardrails scan/v1/guardrails/scan
Run the AI firewall (prompt-injection detection, PII, credit-card, and SSN redaction) on any text, standalone. Add Blunom's safety check to anything, even flows that never hit a model.
Usage/v1/usage
Pull TokenOps usage and cost by key, model, and team into your own dashboards and FinOps tooling.

Scale and events

Batches/v1/batches
OpenAI-compatible bulk execution, governed per key, drained asynchronously for large jobs.
Webhooks/v1/webhooks
Subscribe to events and receive signed, delivered callbacks so your systems react to runs and governance events.

Every endpoint speaks a standard SDK, enforces your firewall and spend policy, and returns a trace id. One key, one base URL, one audit trail across all of it.

One control point, three teams

For the CTO and platform teams

Integrate once, build on any model, and never rewrite to switch providers or add fallback.

For the CISO and IT compliance

Enforce one AI firewall and policy on every model and tool call, with a complete audit trail for every request across the org.

For the CFO and FinOps

Cap spend per key and per team, see cost by model, and stop surprise bills before they happen.

Built for the people who govern AI

For the CISO

Enforce one AI policy on every model and tool call.

One AI firewall and content policy enforced on every model and tool request
Prompt-injection detection and PII, credit-card, and SSN redaction on input and output
Model and action allow-lists, with a fail-closed opt-in for destructive actions
A complete audit trail with a trace id on every request

For the CFO

Cap AI spend and see cost by model, without locking in a vendor.

Hard spend caps per key and per team, enforced before the spend happens
Cost by model across every provider you connect
Bring your own provider keys, so you keep your negotiated pricing

Frequently asked questions

Make Blunom the control point for AI in your business

Book a demo

Blunom Agentic RouterBuilt-in AI Firewall and Real-time TokenOps

Your own AI gateway that routes every model and tool, with governance built in, plus an AI Firewall and real-time TokenOps

AI IS FRAGMENTED

Compatible with what you already run

Models and Providers

Tools

SDKs

Point your SDK at Blunom. Keep your code.

Connect once

Route anything

Govern everything

Everything a control plane for AI should do

Route Any Model

01Drop-in SDK compatibility

02Model routing and automatic fallback

03Hot-swap any model, no code change

04Bring your own keys (BYOK): use your existing provider accounts and discounts

Govern Every Call

05Built-in AI firewall

06TokenOps spend governance: spending caps per key and team, with cost by model

07Model and action allow-lists

08Unified observability and audit

Run Tools and Knowledge

09Cross-provider tool catalog

10Governed knowledge (opt-in RAG): let models search approved company documents, only if you turn it on

One API. Every primitive.

Inference, in the SDKs you already use

Agents, not just models

Tools and actions

Knowledge and memory

Governance as a primitive

Scale and events

One control point, three teams

For the CTO and platform teams

For the CISO and IT compliance

For the CFO and FinOps

Built for the people who govern AI

For the CISO

For the CFO

Frequently asked questions

What is the Blunom Agentic Router?

Is this an AI gateway?

Do I have to rewrite my code to use it?

What API endpoints does the Agentic Router expose?

Can I run agents, tools, and embeddings through the same key?

Can I switch or add models without changing my application?

Does it use my own provider accounts?

How does it keep AI usage secure and compliant?

Can it call tools and business applications, not just models?

Make Blunom the control point for AI in your business