Sécurité Eden AI

La sécurité à chaque niveau de votre gateway IA

Eden AI aide les équipes à accéder aux modèles d’IA via une gateway sécurisée, avec une sécurité certifiée, le Zero Data Retention, des endpoints régionalisés, un contrôle par provider et des guardrails intégrés.

Parler à notre équipe

Commencer à construire en toute sécurité

Certifié SOC 2 • Certifié ISO 27001 • Prêt pour le RGPD

Monitoring · Cost overview

Platform

Models

Monitoring

API keys

Routing

Organization

Members

Guardrails

Billing

Spend

$4,500

−39% vs. direct prices

Saved this month

$2,870

routing · cache · fallbacks

Cache hits

23%

441k requests served at $0.00

Fallbacks

142

every one billed once

Daily spend · direct provider prices vs. with Eden AI

Direct provider pricesWith Eden AI

Where the savings came from

Cheaper providers

$1,842

Response cache

$612

Prompt caching

$318

Fallbacks billed once

$98

Recent requests

Time	Model asked	Served by	Latency	Cost
14:02:11	gpt-5.4-nano	openai · eu-west	143 ms	$0.000202
14:02:09	claude-sonnet-5-5	anthropic	1.2 s	$0.001784
14:02:08	gpt-5.4-nano	cache hit	2 ms	$0.000000
14:02:05	mistral-large-latest	mistral fallback	890 ms	$0.001400
14:02:01	gemini-3.8-flash	google	310 ms	$0.000669

Spending cap · prod-support

$640 / $1,000

64%

Resets monthly · key stops at $0

Built to shrink the bill, not the quality.

Lower token spend

Up to 40%

AI models

500+

Providers

50+

Markup on provider prices

Certifié SOC 2

Certifié ISO 27001

Budget caps per API key

Points de terminaison régionaux

Sécurité certifiée

Six levers, all at the gateway

Most AI spend goes to habits: one frontier model for everything, retries that bill twice, prompts repeated word for word, and nobody watching the meter. Each lever below works without changing your application code.

Cost-based smart routing

Leave the provider off the model name and Eden AI sorts providers by cost, sending more traffic to the cheaper ones. Sort by speed or latency when that matters more.

Provider	Latency	$ / M in	Traffic
deepinfra	410 ms	$0.09	52%
azure · eu	180 ms	$0.15	31%
openai	160 ms	$0.15	17%

routing.sort = cost · cheaper providers receive more traffic

The right model for each step

Agents spend most turns reading, searching and verifying. Send those to small models and keep frontier models for the hard steps.

Explore · 73%

small model

Verify · 13%

mid model

Implement · 14%

frontier

Typical coding-agent split: most turns read, search and test. Only edits need the frontier model.

Fallbacks billed once

Declare up to three fallback models or providers. A failed attempt is never billed; you pay only for the attempt that produced a response.

openai/gpt-6-sol · timeout

not billed

anthropic/claude-sonnet-5-5 · 200 OK

$0.0018

"fallbacks": ["anthropic/claude-sonnet-5-5"]

Response caching

Identical requests are served from cache at no additional cost. On by default, toggled per project.

Cache hit rate

23%

Saved

$612

this month

441k requests answered from cache · no provider call, no charge

Prompt caching passthrough

Reuse stable prompt prefixes with OpenAI, Anthropic, Bedrock, Gemini and DeepSeek at the provider's discounted cache-read rate.

Input tokens	Rate	Cost
Cached prefix · 1.2M	0.1×	$0.36
Fresh · 0.4M	1×	$1.20

The cost field already includes the provider's cache-read price

Caps per key, cost per call

Give each API key a budget that resets daily, weekly or monthly, and read the cost of every request in its response.

Key · prod-support

$640 / $1,000

Resets monthly · stops at $0

Returned on every response

"cost": 0.000202

In US dollars, for that single request

Contrôle de la gateway IA

Same quality, 82% less per request

We ran the same five task types through a single frontier model and through Smart Routing, graded both, and compared the bill. The 82% below is the result measured in that benchmark, on a controlled workload. The "up to 40%" in the page title is the conservative estimate we give for real production traffic, which mixes simple and complex requests; how close you land to the benchmark depends on how much of your traffic a cheaper model can serve.

Monthly cost at 10,000 requests a day

Cost per request

$0.0019

vs. $0.0110 · −82%

Quality score

9.40 / 10

vs. 9.48 with the frontier model

Same prompts, same grading

5 task types

summaries, extraction, Q&A, classification, drafting

Obtenir votre clé API

Lire la documentation

Customer stories

Teams that compare before they commit

We use Eden AI because of its standard interface that connects into various AI providers, so we can test & compare the accuracy and manage vendor risks down the road.

Jan Vosecky

Head of Product @ Re-Hub

I'm using Eden AI for parsing identity documents, it works very well and is much more cost-effective than the previous solution I was using. With Eden AI, I'm billed per request, and not only is it much more affordable, but also I can have backup OCR providers for improved reliability.

Jose Vilches

CEO @ SIM Store

Savings estimate

What would it mean for your bill?

Move the sliders to your own numbers. The estimate uses only two levers, cheaper models on routable traffic and the response cache, so fallbacks and prompt caching are upside.

Your workload

Monthly AI spend today

Requests a smaller model could handle just as well

The easy ones: classification, extraction, short summaries, routine replies. Frontier models keep the hard turns.

Requests that come back word for word (cache hits)

Identical prompts served from cache at no cost. Common with FAQs, templates and fixed instructions.

Estimated monthly savings

Estimate only. The low end assumes cheaper models cost 35% less on routable traffic and 80% of repeated requests hit the cache; the high end assumes 60% and 100%. Your actual figure depends on your prompts and traffic mix.

Get your API key

How it works

Three steps to a smaller invoice

Connect

Point your SDK at Eden AI

Change the base URL of your OpenAI-compatible client and keep your code. Prompts, tools, streaming and structured outputs keep working as before.

Route

Pick a strategy, add fallbacks

Sort providers by cost, speed or latency, restrict the pool to approved providers or the EU region, and list fallback models. It is a few fields in the request, not a rewrite.

Measure

Set your quality bar, then cut the cost

Compare models on accuracy, latency and price before you move traffic, then read the cost field on every response and cap each key.

Garde-fous intégrés

Spend less without giving up quality

Three places where the gateway does the work: how each key routes, what the catalog lets you compare before you move traffic, and how little has to change in your code.

01 · Routing

A routing strategy per key, set once

Sort providers by cost, speed or latency, narrow the pool to approved providers, pin regulated traffic to Europe with the @eu suffix, keep a conversation on one provider with sticky routing, and list fallbacks. Configure it in the dashboard or per request.

Explore Smart Routing

Routing strategy · prod-support

Sort providers by

costspeedlatencyexact

Allowed providers

openaiazuremistraldeepinfra+ add

Allow fallbacks

Sticky routing per conversation

Region

Any regionEU only · @eu

Fallback models

mistral/mistral-small-latestgoogle/gemini-3.8-flash

Save routing strategyPreview cost impact

02 · Comparison

Compare price, latency and region before you commit

The model catalog lists every model with its context window, input and output price per million tokens, capabilities and hosting region. The same task can cost 2 to 3x more on one provider than on another, so choose with the numbers in front of you.

Browse models

Eden AI model catalog showing price, latency and region for each model

03 · Integration

Keep your SDK, and your own keys if you want

Eden AI is a drop-in for the OpenAI SDK. With Bring Your Own Keys you keep your negotiated provider prices while Eden AI still handles routing, fallbacks, caching and cost monitoring.

Read the docs

Code console calling Eden AI with an OpenAI-compatible client

Comparative table

Calling providers directly vs. Eden AI

Direct provider integrations

Eden AI

One frontier model for every request, including the simple ones.

Cost-based routing sends each request to the cheapest capable provider.

Prices checked by hand, once, when you integrate.

Live price, latency and region comparison across 50+ providers.

Client-side retries that bill for every attempt.

Fallbacks across models and providers, billed only for the attempt that answers.

Repeated prompts paid in full every time.

Response caching at no extra cost and provider prompt-cache passthrough.

Spend discovered on the invoice at month end.

Cost on every response, budgets per key, one consolidated invoice.

Developer experience

Switch the base URL, keep your code

Eden AI is OpenAI-compatible. Routing and fallbacks are fields on the request, and the response tells you which provider answered and what it cost.

PYTHON · OPENAI SDK

# pip install openai
from openai import OpenAI

client = OpenAI(
    base_url="https://api.edenai.run/v3",
    api_key=EDENAI_API_KEY,
)

response = client.chat.completions.create(
    model="gpt-5.4-nano",        # no provider prefix: Eden AI picks the provider
    messages=[{"role": "user", "content": "Summarize this ticket: ..."}],
    extra_body={
        "routing": {"sort": "cost", "allow_fallbacks": True},
        "fallbacks": ["mistral/mistral-small-latest"],
    },
)
print(response.choices[0].message.content)

RESPONSE · COST INCLUDED

{
  "id": "chatcmpl-8f1c…",
  "model": "openai/gpt-5.4-nano",   // the provider that served the request
  "choices": [
    { "message": { "role": "assistant", "content": "The customer reports…" } }
  ],
  "usage": {
    "prompt_tokens": 412,
    "completion_tokens": 96,
    "total_tokens": 508
  },
  "cost": 0.000202                   // USD, on every response
}