Agent Lab · Enterprise AI gateway

One key for every model,
every call priced before it runs.

Cloud or on-prem models. Point Claude Code or Codex at one URL and see who spent what.

The basics

What is an AI gateway?

An AI gateway (or LLM gateway) is the checkpoint between a company and its AI models: every tool calls models through it, and it picks the model, enforces budgets and records each call.

Models: Claude and GPT in the cloud, or your on-prem models
APIs: OpenAI and Anthropic formats
Tools: Claude Code, Codex, OpenAI / Anthropic SDKs
Model choice: Name one, or let auto pick by difficulty
Limits: Workspace monthly budget, team ceilings
Records: Key, team, model and cost per call

Why

A model key per team is a bill nobody controls

No ceiling

A runaway script shows up at month end.

No team split

One shared key, no idea who spent it.

No trail

A total, but not which tool.

Models everywhere

Switching means editing every tool.

How it works

Every call takes the same path

Check the key
Each key counts against a team.
Pick the model
Named is used as named; auto picks.
Hold the budget
Over the limit, it is not sent.
Record
Every cost stays on record.

Auto model choice

Cheap models for simple work, strong ones for hard

Task	Model	For example
Multi-step	Stronger	Edit several files
Analysis, coding	Standard	Explain a function
Chat, translation, extraction	Cheapest	Pull fields from a doc
Could not tell	Standard	Never quietly cheaper

After 80% of the monthly budget, hard tasks get the standard model too. CLIs turn auto on with one header, X-AgentLab-Routing: auto.

Budgets and limits

Priced before it is sent, not at month end

Monthly budget

Plus a 6-hour rolling limit.

Team ceilings

A USD ceiling for each team.

Hold, then settle

Concurrent calls cannot overspend.

80% and 100% alerts

The console tells the owner.

Connect

Connect Claude Code, Codex and the OpenAI SDK

Anthropic Messages

Claude Code

The last header turns on auto.

export ANTHROPIC_BASE_URL=https://api.agentlab.cresclab.com
export ANTHROPIC_AUTH_TOKEN=<your API key>
export ANTHROPIC_CUSTOM_HEADERS="X-AgentLab-Routing: auto"
claude

OpenAI Responses

Codex

Goes in ~/.codex/config.toml.

model_provider = "agentlab"

[model_providers.agentlab]
name = "Agent Lab"
base_url = "https://api.agentlab.cresclab.com/v1"
wire_api = "responses"
env_key = "AGENTLAB_API_KEY"
http_headers = { "X-AgentLab-Routing" = "auto" }

Chat Completions

OpenAI SDK

The Anthropic SDK also just changes base_url.

client = OpenAI(
    base_url="https://api.agentlab.cresclab.com/v1",
    api_key=os.environ["AGENTLAB_API_KEY"],
)
client.chat.completions.create(model="auto", messages=[...])

Once access is on, the console writes these settings for you.

Records and audit

Every call can be traced

Request log

Key, team, model, cost.

Spend report

Down to a single key.

Keys stored as hashes

Revocable and audited.

Dropped calls settle

Automatically after 15 minutes.

Compare

How it compares with other AI gateways

	Hosting	Budget enforcement	Auto model choice	Best for
Agent Lab AI Gateway	Run by us; cloud and on-prem models	Held before sending	Auto by difficulty	Companies that do not want to run it
LiteLLM	Open source, self-hosted	Reserved before sending	Auto Router (beta)	Engineering teams that self-host
Portkey (now Prisma AIRS AI Gateway)	SaaS, hybrid, self-hosted	Per-key limits	Rule-based	Logs and guardrails first
Cloudflare AI Gateway	Run by Cloudflare; core free	Charged after completion	Auto Router	Teams already on Cloudflare
Kong / Azure APIM	Extends existing API management	Kong: next request; Azure: tokens only	Load balancing	Teams already on them
TrueFoundry	SaaS; enterprise self-host	Per user or model	Weight, latency	Teams that also host models

From each vendor's docs, checked 2026-10-21: LiteLLM, Portkey, Cloudflare, Kong, Azure, TrueFoundry.

FAQ

AI gateway questions

What is an AI gateway?

An AI gateway (LLM gateway) is the checkpoint between a company and its AI models. Every tool calls models through it, and it picks the model, enforces budgets and records each call.

Can it connect on-prem models?

Yes. Any on-prem model with an OpenAI-compatible API (for example vLLM or Ollama) connects alongside cloud Claude and GPT, under the same budgets and records.

How is it different from LiteLLM?

LiteLLM is open source and you host and run it. Agent Lab is run for you and works once access is on. Both hold the budget before a call is sent.

How is it different from Cloudflare AI Gateway?

Cloudflare's core features are free, but cost is recorded after a request completes. Agent Lab holds it before sending, so concurrent calls cannot overspend together.

Does auto make answers worse?

Only simple tasks get a cheap model; when unsure it uses the standard one. A named model is never swapped.

How much will it save?

It depends on your tasks, so there is no fixed percentage. The report shows the average cost per completed request; compare before and after.

Request access,
connect one tool first

Which models and tools you use
Rough monthly usage
Create a key and connect

Already have access? Sign in to the console.

One key for every model,every call priced before it runs.