AcruxCore

What is AcruxCore?

AcruxCore is an open-source LLMOps platform that sits between your app and model providers. Teams version prompts, route calls through one OpenAI-compatible gateway, and trace every request with token, latency, and cost data. Built-in evaluations score prompt changes against datasets, so you can promote a better version without redeploying your app.

Top Features:

• Prompt versioning: move a production alias between versions without redeploying.
• LLM gateway: route calls to any provider with fallbacks, caching, and budgets.
• Evaluation: score prompt versions on datasets and let the optimizer draft rewrites.

Use Cases:

• Debugging bad answers: open the trace behind a reply and find its prompt.
• Cost control: set budgets, rate limits, and caching per key.
• Agent monitoring: follow a CrewAI or OpenAI Agents SDK run span by span.

Who Can Use AcruxCore?

• AI engineers: ship and change LLM features without a full deploy cycle.
• Platform teams: put one gateway in front of every model provider.
• Regulated companies: self-host the stack so prompts, keys, and data stay inside.

Pricing

• Self-hosted (free): the full Apache 2.0 platform on your own infrastructure.
• Hosted Beta (free): every feature, using your own provider keys with no markup.
• Enterprise (contact sales): custom data retention, named support contacts, and security reviews.

Pros and Cons

Pros:

• No token markup: model spend stays exactly what your providers already charge.
• Fully open source: the API, gateway, dashboard, and SDKs are public.
• Modular adoption: use prompts or tracing without routing traffic through the gateway.

Cons:

• Beta status: hosted pricing after the beta has not been announced.
• Developer focus: setup needs TypeScript or Python SDK work, not a no-code editor.
• Young ecosystem: fewer community guides exist than for older LLMOps tools.

FAQs:

1) Is AcruxCore free?

Self-hosting is always free, and the hosted platform costs nothing during the current beta.

2) Do I need a new OpenAI client?

No, the gateway is OpenAI-compatible, so changing the base URL keeps your code working.

3) Which SDKs are available?

There are TypeScript and Python SDKs, plus a plain REST API for other languages.

4) Does it support OpenTelemetry?

Yes, you can export your own spans over OpenTelemetry instead of routing through the gateway.

5) Does it mark up model tokens?

No, you connect your own provider keys and pay those providers directly with no markup.