Multi-provider access
Give every team one OpenAI-compatible endpoint and virtual API keys. Add, swap, or self-host providers behind the gateway with zero changes to application code.
Governed model request
LLM Gateway
Govern every model call your teams make
Providers multiply, keys leak, and spend surprises arrive at the end of the month. Odock puts a governed LLM gateway between your apps and every model provider, enforcing access, guardrails, budgets, and audit records before any completion is generated.
One governed path for every model call.
llm-gateway.request
LLM Gateway
>POST /v1/chat/completionsapp runtime
01Caller authverified
02Model grants...
03Guardrail scan...
04Budget hold...
05Route provider...
06Usage logged...
Providers
OpenAI, Anthropic, Azure, self-hosted
Controls
Keys, budgets, guardrails, audit
What is an LLM gateway?
One controlled endpoint between your apps and every model provider
An LLM gateway is a control plane between your applications and model providers like OpenAI, Anthropic, Azure, or self-hosted models. Instead of every team holding raw provider keys and calling APIs directly, model traffic flows through one endpoint that authenticates the caller, applies guardrails, reserves budget, routes to an approved provider, and records the outcome. It becomes a governance platform when it goes beyond routing: access grants per team, budgets held before the call fires, prompt injection and data leakage scanning, and audit-ready usage records. That is what Odock is built for, and the same plane also governs agent tool traffic through the MCP gateway.
Where teams use it
Cost control that blocks, not reports
Set budgets and quotas per key, team, or project. Odock reserves spend before the call executes, so a runaway agent stops at the limit instead of at the invoice.
Security and compliance at the gateway
Scan prompts for injection and data leakage, mask sensitive fields, enforce provider data policies, and produce audit records for every model decision.
Keep exploring
LLM request lifecycle
Every model call follows a governed path
No request reaches a provider until it passes auth, access, inspection, and cost controls. Every outcome is recorded with tokens, latency, and cost.
01
Authenticate
Validate the virtual API key.
02
Authorize
Confirm model and provider grants.
03
Inspect & enforce
Run guardrails on prompt and context.
04
Reserve spend
Check budgets and quotas before execution.
05
Route
Send to an approved provider with failover.
06
Record outcome
Log tokens, latency, status, and cost.
Blocked request exampleDenied by governance
{
"apiKey": "vk_team_marketing",
"model": "gpt-5.2",
"method": "chat.completions",
"reason": "budget_exceeded",
"status": 402
}Why it matters
Model calls are spend, data, and compliance events
Every completion moves money, may carry sensitive data, and may need to be explained to an auditor later. An LLM gateway gives platform, security, and finance teams one enforcement point for all three.
Control which models teams use
Allowlist approved providers and models per team or key, block deprecated or non-compliant models, and roll out new models without touching application code.
Enforce policy before execution
Authenticate every caller, scan prompts for injection and data leakage, and reject requests that fail security, budget, or compliance rules before any provider call is made.
Attribute every token
Tie every request to a key, team, and user with tokens, latency, and cost. Give finance chargeback data and give auditors the records they ask for.
What you configure
A complete LLM control surface, not just a proxy
Odock covers everything platform teams need to run LLM traffic in production: provider registration, model access grants, guardrails, pricing, budgets, and usage records your finance and security teams can actually read.
LLM views
Selected view
Register providers once and expose approved models through one OpenAI-compatible endpoint. Review API type, auth config, scope, and enabled status before any team can call them.
providers-&-models.llm
Live
openai-prod
Provider catalog
APIOPENAI_API
AuthBEARER
Scopeorg-wide
Access24 keys granted
anthropic-prod
Provider catalog
APIANTHROPIC_API
AuthBEARER
Scopeorg-wide
Access18 keys granted
azure-openai-eu
Manual setup
APIOPENAI_API
AuthBEARER
Scopeteam:eu-products
Access6 keys granted
bedrock-us
Provider catalog
APIBEDROCK_API
AuthOAUTH2
Scopeteam:platform
Access4 keys granted
mistral-eu
Provider catalog
APIOPENAI_API
AuthBEARER
Scopeteam:eu-products
Access5 keys granted
vllm-selfhosted
Manual setup
APIOPENAI_API
AuthNONE
Scopeteam:research
Access3 keys granted
One endpoint, two ways in
Your apps call Odock. Not the provider
Reach Odock through one unified OpenAI-style endpoint across every provider, or through each provider's native endpoint and SDK when you need provider-specific features. Either way, Odock authenticates the caller, confirms model access, runs guardrails, reserves budget, injects provider credentials, and records the outcome before the provider sees the request.
What the gateway handles
- Unified OpenAI-style API across every provider, so existing SDKs work unchanged.
- Native provider endpoints and SDKs when you need provider-specific features.
- Virtual API key required before any model is reachable.
- Provider credentials injected after governance checks pass.
- Routing and failover across approved providers.
llm-unified-openai.py
UnifiedOpenAI SDK
Method
Language
1# Use Odock's unified endpoint through an OpenAI-compatible SDK2import os3from openai import OpenAI4 5client = OpenAI(6 api_key=os.environ["ODOCK_API_KEY"],7 base_url=os.environ.get("ODOCK_BASE_URL", "https://api.odock.ai/v1"),8)9 10response = client.chat.completions.create(11 model=os.environ.get("ODOCK_MODEL", "claude-sonnet-4-5"),12 messages=[13 {"role": "user", "content": "Explain budget enforcement."}14 ],15 temperature=0.2,16 max_tokens=200,17)18 19print(response.choices[0].message.content)EU AI Act & compliance readiness
Evidence for every model decision
Compliance programs need answers to specific questions: which team used which model, under what policy, with what safeguards? Direct provider access can't answer those questions. Odock can.
- Route every provider through one governed, auditable OpenAI-compatible endpoint
- Enforce prompt injection and data leakage controls outside application code
- Produce evidence for internal AI reviews, vendor assessments, and EU AI Act compliance programs
- Keep LLM traffic and MCP tool traffic inside the same policy and operational boundary
FAQ
Questions teams ask about LLM gateways
These are the recurring questions that come up when teams move from direct provider access to a governed gateway layer.
One endpoint for every model, with governance built in
If your teams call LLMs in production, you need the same control you expect for any critical dependency. Odock governs model and tool traffic from a single request path.
llm-unified-openai.py
# Use Odock's unified endpoint through an OpenAI-compatible SDKimport osfrom openai import OpenAI client = OpenAI( api_key=os.environ["ODOCK_API_KEY"], base_url=os.environ.get("ODOCK_BASE_URL", "https://api.odock.ai/v1"),) response = client.chat.completions.create( model=os.environ.get("ODOCK_MODEL", "claude-sonnet-4-5"), messages=[ {"role": "user", "content": "Explain budget enforcement."} ], temperature=0.2, max_tokens=200,) print(response.choices[0].message.content)