Kodwen

All comparisons

Compare

Kodwen vs LiteLLM

What each one is, how it is priced, where data lives, and when LiteLLM is the better pick.

Last checked:

In short

LiteLLM is an open-source gateway (Python SDK or proxy) that you deploy and run yourself, with a paid enterprise licence. Kodwen is a managed service: nothing to deploy, one model name, billed per verified task in EUR.

When LiteLLM is the better pick

Pick LiteLLM if you need to self-host the gateway (including air-gapped), keep every request inside your own infrastructure, or want an MIT-licensed core you can modify.

Kodwen compared with LiteLLM, checked on October 2, 2026
DimensionKodwenLiteLLMSources
What it is / deploymentManaged gateway (SaaS). Runs on Cloud Run in GCP europe-west1 (Belgium). One public model name: kodwen.Open-source AI gateway: a Python SDK or a self-hosted proxy (default http://0.0.0.0:4000) that calls 100+ providers in the OpenAI format. No hosted offering could be confirmed.
Open sourceNo. Not open source and not self-hostable.Yes: MIT outside enterprise/. Enterprise code needs “a valid BerriAI Enterprise license” in production; SSO is free for up to 5 users.
Wire formatsOpenAI Chat Completions, OpenAI Responses and Anthropic Messages./chat/completions, /responses (with a WebSocket mode), Anthropic /v1/messages, plus embeddings, images, audio, batches, rerank and a2a.
Pricing modelMonthly plans in EUR: Free (1 verified task, 300 Lite + 20 Smart chat messages), Lite €4.99, Kodwen Services €3.99 (your own model keys), Student €5.99 (2 tasks), Pro €19.99 (10 tasks), Builder €49 (28), Max €99 (60), Team €25 per seat (12 per seat). Overage €1.20 per verified task; unverified tasks are billed at raw token cost.Open-source edition “$0 Free forever”. Enterprise: “Pricing depends on your deployment size”, no published prices, 30-day trial; 24/7 SLA support is an additional fee.
Data residencyGateway and database in the EU (europe-west1, Belgium). Model inference region depends on the upstream deployment.Self-hosted: “We run no telemetry when you self-host LiteLLM”; data stays in your infrastructure. Air-gapped deployment supported; EU deployment filter (region_name: "eu").
Routingkodwen routes each request to the model best suited to it among Anthropic's current Claude models and fails over between approved deployments; the x-codai-routed-to header names the model that answered.Strategies: simple-shuffle (default), least-busy, usage-based, latency-based, cost-based, custom; retries, cooldowns, several fallback types; a beta complexity-based auto router.
Prompt cachingOn by default for every request.Passes cache_control through (translated to Bedrock cachePoint); automatic Anthropic caching is “Off by default”; its own response cache (Redis/Valkey, semantic, S3, GCS, memory, disk).
MCPRemote MCP server (Streamable HTTP) with OAuth 2.1.MCP Gateway inside the proxy: Streamable HTTP, SSE and stdio; access control by key, team or org; OAuth 2.0 and SigV4.
Billing currency / invoicesPrices and charges in EUR; invoices issued by the Romanian seller.Enterprise procurement by invoice, AWS Marketplace or Azure Marketplace. Billing currency is not stated on its pages.
Company HQIMS SRL (Interactive Media Solutions S.R.L.), Târgu Jiu, Romania.Legal entity “Berrie AI Incorporated” (YC W23 badge). Its own pages do not state an HQ country.

Where LiteLLM’s own pages disagree

Sources

Facts from each vendor’s public pages on October 2, 2026; tell us at hello@codai.ro if something changed.