Open Source
AISIX
by API7.ai
An open-source gateway written in Rust, from the original creators of Apache APISIX, that puts one OpenAI-compatible and Anthropic-compatible API in front of OpenAI, Anthropic, Gemini, Bedrock, Azure OpenAI and other providers.
Our verdict
An open-source gateway written in Rust, from the original creators of Apache APISIX, that puts one OpenAI-compatible and Anthropic-compatible API in front of OpenAI, Anthropic, Gemini, Bedrock, Azure OpenAI and other providers.
Best for: Platform teams that want a single-binary gateway with routing, failover, rate and token limits, guardrails and caching, configured from files.
Choose something else if: you need budgets, a dashboard or encrypted key storage without paying. Those come with AISIX Enterprise or AISIX Cloud.
Source review
Using AISIX in practice
Where it runs
The gateway runs in your environment. The paid Cloud and Enterprise control planes manage it but do not carry live model traffic.
How to install
Needs Docker and a provider key. Write config.yaml and resources.yaml as shown in the README, export your provider key and a caller key, then run the container. Clients call http://localhost:3000/v1 with the caller key.
Install commands
docker run -d --name aisix \
-v "$(pwd)/config.yaml:/etc/aisix/config.yaml:ro" \
-v "$(pwd)/resources.yaml:/etc/aisix/resources.yaml:ro" \
-e OPENAI_API_KEY -e CALLER_API_KEY \
-p 3000:3000 -p 127.0.0.1:9090:9090 \
ghcr.io/api7/aisix:latestAs published in the linked sources. Not run by us.
What it costs
The gateway is free under the Apache 2.0 licence, and you pay providers directly with no token markup. Optional AISIX Cloud management is free for 100,000 recorded requests a month, then US$149 a month for Team, with Enterprise quoted.
Prices and plans change. Check the linked sources before committing.
Why consider it
- Runs in your environment and calls providers directly, so live traffic never passes through API7.
- Also acts as a gateway for MCP servers and agent-to-agent traffic, with metrics and telemetry export.
What to watch for
- Young and fast-moving: first released in July 2026, with breaking changes in recent versions.
- The metrics and status port, 9090, has no authentication. Keep it off public networks.
- The open-source edition has rate and token limits but no budgets or spend caps.
A first task to try
Run the quickstart with one provider key and one caller key, send a chat request to port 3000, then check the access log entry for its model and token counts.
Suggested exercise, not an installation test. We reviewed the linked documentation; no host compatibility is implied.
Other options to understand
Related tools can solve different parts of the same job. Compare fit before treating them as replacements.
Key facts
What we have checked
- GitHub stars
- 177
Each value shows when we last confirmed it and where it came from. Values we have not confirmed are left out, never guessed.
Sources
Where this comes from
- Repository and READMEOur review
- Gateway quickstartOur review
- PricingOur review
- github api via nightly crawl, recovered from git history 4c14e39bad662f81cc532f907f2dbc9656782a39Registry
A source shows what was published when we checked it. It does not certify security, performance or a working installation. Read our method.
Changelog
What changed
No changes recorded yet. Releases, price changes and deprecations appear here, with their dates, once we track them for this record.
Get the free Developer’s Field Guide
A 27-page field guide to the AI coding workflow with Claude. Claude Code, MCP servers, the prompt patterns that work, and what to delegate. Free.
Enter your work email. We send it straight over, plus a few short notes worth knowing. Unsubscribe any time.