Open source / self-hosted 10 services

Open-source self-hosted gateways and deployment tooling — run your own relay layer and keep the data entirely under your control. Suited to technical teams with ops capability.

SambaNova

sambanova.ai
Verified Stability: Very stable GPT DeepSeek Global / overseas

The site itself puts it this way: Discover - the complete AI platform delivering the fastest AI…

Score 8.1 Uptime 100% Latency 608ms Models 6
Open source Stability: Poor Claude GPT Gemini Global / overseas

GitHub (helicone/helicone) is an overseas AI API relay and aggregation service. Model coverage s…

Score 7.0 Uptime 70.49% Latency 1077ms
Open source Stability: Poor Claude GPT Gemini Global / overseas

A globally hosted relay that fronts a range of AI model APIs. Claude, GPT, Gemini, DeepSeek and…

Score 6.8 Uptime 70.49% Latency 1100ms
Open source Stability: Poor Claude GPT Global / overseas

An overseas AI API relay and aggregation service. Model coverage spans Claude, GPT. You supply t…

Score 6.6 Uptime 70.49% Latency 1463ms

GitHub (berriai/litellm)

github.com
Open source Stability: Poor Claude GPT Gemini Global / overseas

An overseas AI API relay and aggregation service. Model coverage spans Claude, GPT, Gemini. The…

Score 6.6 Uptime 70.49% Latency 1107ms

GitHub (labring/FastGPT)

github.com
Pending review Stability: Poor GPT Global / overseas

A globally hosted relay that fronts a range of AI model APIs. It covers the GPT families. The st…

Score 6.4 Uptime 75.41% Latency 1032ms
Pending review Stability: Poor Claude Global / overseas

GitHub (Portkey-Ai/gateway) is an overseas AI API relay and aggregation service. Claude and othe…

Score 6.4 Uptime 72.13% Latency 1084ms

GitHub (langgenius/dify)

github.com
Pending review Stability: Poor Claude Self-hosted

A self-hosted API gateway project: you run it, you keep the data. Claude and other families are…

Score 6.4 Uptime 72.13% Latency 1136ms

dify.ai

dify.ai
Pending review Stability: Very stable Global / overseas

Per the service's own site: Dify is the platform for production-ready agentic workflows. The sta…

Score 6.2 Uptime 100% Latency 931ms

About this category: Open source / self-hosted

This category is not relays at all, but open-source projects you can use to run a relay of your own — One API, New API, LiteLLM and similar. What they share: open code, self-deployable, and data entirely under your own control.

When is self-hosting worth it? The test is fairly clear: the data is genuinely sensitive and cannot pass through a third party; the call volume is large enough to amortise the operational cost; and someone on the team can take on deployment and maintenance. If any one of the three is missing, a ready-made service is usually the better deal.

The cost of self-hosting is not just a server. You also have to account for where the upstream keys come from (you connect to each vendor's open platform yourself), high-availability design (a single-machine deployment stops everything when it fails), and ongoing maintenance (upstream API changes have to be tracked). Once all of that is counted, "self-hosting is cheaper" usually does not hold — what you buy is control, not savings.

A pragmatic middle path is a self-hosted gateway with official upstreams: use a tool such as One API for unified authentication, quotas and logs, and connect directly to each vendor's official platform. That keeps control of your data and accounting without having to run the forwarding path yourself.

How to choose within this category

The picks below are computed live from the current measurements, not fixed conclusions — change the data and the picks change with it. Each states its basis, and you can open the detail page to check it yourself.

  • Most stable:Highest current uptime. Fewer samples means more distortion, so we state the source (public dataset / our own measurement) and you can check it on the detail page.
  • Fastest:Ordered by the home-page response time of the most recent probe. A single measurement is noisy, so treat it as a range rather than an absolute.
  • Broadest model coverage:This uses the model count our endpoint probe actually detected, not the support list the service advertises — the two are measured differently, and the advertised figure is usually the more optimistic one.

Key parameters side by side

The top 10 services by overall score in this category. 0 of them are confirmed to expose a /v1/models endpoint — that figure comes from probing, not from the services' own claims.

ServiceRegion UptimeLatencyModels PaymentAPI
SambaNova
sambanova.ai
Global / overseas 100% 608 ms 6 — Unconfirmed
GitHub (helicone/helicone)
github.com
Global / overseas 70.49% 1077 ms — Credit card Unconfirmed
GitHub (1Panel-dev/MaxKB)
github.com
Global / overseas 70.49% 1100 ms — — Unconfirmed
GitHub (VoltAgent/voltagent)
github.com
Global / overseas 70.49% 1463 ms — — Unconfirmed
GitHub (berriai/litellm)
github.com
Global / overseas 70.49% 1107 ms — — Unconfirmed
GitHub (Portkey-Ai/gateway)
github.com
Global / overseas 72.13% 1084 ms — — Unconfirmed
GitHub (labring/FastGPT)
github.com
Global / overseas 75.41% 1032 ms — — Unconfirmed
GitHub (langgenius/dify)
github.com
Self-hosted 72.13% 1136 ms — — Unconfirmed
The Fastest Enterprise AI Gateway
langdb.ai
Global / overseas 100% 994 ms — — Unconfirmed
dify.ai
dify.ai
Global / overseas 100% 931 ms — Credit card Unconfirmed

Frequently asked questions

The questions we see most often about this category. The guidance applies to the category as a whole, not to any single service.

Is self-hosting cheaper than using a ready-made service?

Usually not, unless your volume is large. What you save is the rate markup; what you add is server cost, the labour for deployment and maintenance, and the work of high-availability design. At modest monthly volume, a ready-made service's markup is normally less than the hidden cost of self-hosting.

What is the difference between One API and New API?

New API is a fork that continues active development on top of One API, with a faster-moving feature set and quicker support for newer protocols. Choose on maintenance activity and the range of upstreams you need — both provide the core ability to unify multiple upstreams behind one endpoint.

Where do the upstream keys come from once I self-host?

You register on each vendor's open platform and obtain them yourself. That means dealing with each vendor's identity requirements, payment methods and quota limits — work that is routinely underestimated. If avoiding the markup is your only reason, price this part first.

Does a self-hosted gateway guarantee no downgrading and no substitution?

Yes — and that is the most concrete value of self-hosting: there is no third party in the path, so the model you request is the model you reach. The price is carrying upstream stability yourself: when the official platform throttles you, there is no alternative channel.

Is self-hosting suitable for a small team?

Not if nobody can maintain it continuously. Self-hosting means tracking upstream API changes, handling certificates and domains, and running backups and monitoring. A reasonable compromise is to start with a ready-made service and re-evaluate once both your volume and your data sensitivity pass a threshold.

Keep exploring this category

This page shows you what exists; the pages below show you who is more stable, faster and cheaper. Read together, they tell you more than either alone.