//API PLATFORM

One sovereign API,
compatible with what you already use

You already wrote the code. Point it at AVELIN instead and it just works: the same OpenAI or Anthropic calls, our own models underneath. Run it in our cloud for 50% less than OpenAI and Anthropic, or on your own hardware for up to 10× less. Two lines change. Nothing else has to.

Base URLhttps://api.fi.avelin.ai

OpenAI + Anthropic compatible9 models · 3 families~90% routing precisionSovereign · on-prem · air-gapped

Read the Docs Book a Demo

//ARCHITECTURE

How a request flows

Every call runs through the same governed path: classified, routed to the right specialist, optionally fused, and streamed back. Routing adds ~50ms and the whole flow returns in ~2.6s median.

Semantic Router~50ms

Classifies the task type and selects the optimal model family.

Cross-Model MoE Classifier

Picks the specialist tier by cost, latency, quality, and availability.

Model Fusion (optional)

Queries several experts in parallel and synthesizes one stronger answer.

Confidence Cascading

Escalates to a premium tier only when the confidence score is low.

Response Transformation

Normalizes the response format and streams it back to your code.

Continuous Evaluation48h new models

Runs in the background: re-benchmarks providers and updates routing weights.

99.97%

uptime, 12-mo avg

0.3s

avg failover

~50ms

routing overhead

17%

fewer hallucinations

Full request flow in the docs

//DROP-IN COMPATIBLE

Keep the code you already have.

AVELIN speaks the OpenAI and Anthropic wire formats, so your existing stack works as-is: no rewrite, no retraining your team, no lock-in. Migrating means swapping the base URL and the key. Everything else stays the same call: models, tools, and streaming included.

OpenAI-compatible

/v1/chat/completions

Best for OpenAI SDKs and most third-party tools.

Anthropic-compatible

/v1/messages

Best for Anthropic SDKs and fine-grained thinking control.

openai_client.py

  client = OpenAI(-     base_url="https://api.openai.com/v1",+     base_url="https://api.fi.avelin.ai/v1",-     api_key="sk-...",+     api_key="sk-ave...xxxx",  )

Anthropic surface: base URL https://api.fi.avelin.ai

//FOR DEVELOPERS

Everything a developer needs, in one place.

Point your code at AVELIN and it just works. Here is every endpoint with a copy-paste example. Authentication is a single bearer token, and full SDK setup, streaming, and error handling live in the docs.

Endpoint	Method	Purpose
`/v1/chat/completions`	POST	Chat & reasoning (OpenAI-compatible)
`/v1/messages`	POST	Chat with thinking control (Anthropic-compatible)
`/v1/models`	GET	List models available to your key
`/v1/embeddings`	POST	Vector embeddings (bge-m3)
`/v1/audio/transcriptions`	POST	Speech-to-text (whisper-large-v3)
`/v1/images/generations`	POST	Image generation (avelin-imagegen)
`/v2/scrape · search · crawl`	POST	Web Intelligence (Firecrawl-compatible)

Full API reference

Streaming (SSE)Tool & function callingReasoning & thinking controlAutomatic retry & failover

// OpenAI-compatible

Chat completions

Call /v1/chat/completions exactly as you would OpenAI. Pick a model by capability tier; the Cross-Model MoE handles routing for you.

// OpenAI-compatible

Chat completions

Call /v1/chat/completions exactly as you would OpenAI. Pick a model by capability tier; the Cross-Model MoE handles routing for you.

chat_completions.sh

curl https://api.fi.avelin.ai/v1/chat/completions \
  -H "Authorization: Bearer $AVELIN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "avelin-pro",
    "messages": [
      { "role": "user", "content": "Summarize the benefits of vendor-neutral AI." }
    ]
  }'

// Anthropic-compatible

Messages with thinking control

Call /v1/messages with the Anthropic wire format. Extended thinking is on by default on supported models and streams separately from the answer.

// Anthropic-compatible

Messages with thinking control

Call /v1/messages with the Anthropic wire format. Extended thinking is on by default on supported models and streams separately from the answer.

messages.sh

curl https://api.fi.avelin.ai/v1/messages \
  -H "Authorization: Bearer $AVELIN_API_KEY" \
  -H "Content-Type: application/json" \
  -H "anthropic-version: 2023-06-01" \
  -d '{
    "model": "avelin-coding",
    "max_tokens": 1024,
    "messages": [
      { "role": "user", "content": "Write a Python function to debounce calls." }
    ]
  }'

// Tool calling

Function & tool use

Both surfaces support tool calling. The model decides when to call your function; agentic models are tuned for multi-step tool use.

// Tool calling

Function & tool use

Both surfaces support tool calling. The model decides when to call your function; agentic models are tuned for multi-step tool use.

tools.json

{
  "model": "avelin-agentic",
  "messages": [
    { "role": "user", "content": "What's the weather in Abu Dhabi?" }
  ],
  "tools": [
    {
      "type": "function",
      "function": {
        "name": "get_weather",
        "description": "Get current weather for a city",
        "parameters": {
          "type": "object",
          "properties": { "city": { "type": "string" } },
          "required": ["city"]
        }
      }
    }
  ],
  "tool_choice": "auto"
}

//CROSS-MODEL MoE

Trained intelligence, not a gateway

Most APIs hand your request to one model and hope it fits. Ours doesn't. A classifier trained on a year of real traffic reads what you send, routes it to the best of our nine specialists, and blends their answers when that makes the result stronger. It gets the call right about 90% of the time. This is a trained model, not a router dressed up as one.

	API aggregator / gateway	AVELIN Cross-Model MoE
Architecture	Request forwarding	Trained MoE meta-model
Intelligence	None: passes through to providers	Synthesized from experts fused with learned gating
Output quality	Limited to a single model	Exceeds any individual model
Improvement	Dependent on upstream changes	Self-improving within 48 hours
Deployment	Cloud-only: data leaves the perimeter	Sovereign: runs on your GPUs

Cost advantage: 50% cheaper than OpenAI and Anthropic in our cloud, and up to 10× cheaper when you run it on your own hardware. Prompt caching cuts the cost of repeated prompt prefixes by up to 90%.

Model ID	Family	Profile	AA Index	Context	Capability
`avelin-ultra`	Intelligence	Deep Reasoning	55	256K	Highest depth for strategic memos, risk analysis, and synthesizing ambiguous data.
`avelin-pro`	Intelligence	Operational Efficiency	53	256K	High throughput for daily automation, summarization, and standard business logic.
`avelin-fast`	Intelligence	Real-Time Execution	47	256K	The fastest tier for high-volume chat, classification, and extraction. Auto-routes to vision when a request contains images.
`avelin-coding-architect`	Coding	Architecture Design	52	1M	System-level design decisions, long-horizon refactors, whole-repo analysis.
`avelin-coding-plus`	Coding	Production Coding	49	1M	Multi-file refactors, latency-sensitive CI pipelines, high-volume agentic loops.
`avelin-coding`	Coding	Standard Coding	42	1M	Day-to-day development tasks, bug fixing, code review, testing.
`avelin-agentic-high`	Agentic	Complex Agents	67	1M	Long-running agents, security-sensitive workflows, complex orchestration.
`avelin-agentic`	Agentic	Standard Agents	67	200K	Tool-calling workflows, task decomposition, multi-step execution.
`avelin-agentic-fast`	Agentic	High-Volume Agents	38	256K	Agent swarms, bulk classification, cost-sensitive agentic backends.

AA Index scores from the Artificial Analysis Index v4.x benchmark. Higher is stronger.

Model catalog

//DEPLOYMENT & SECURITY

Run it where your data lives.

Some data simply can't leave the building. So AVELIN doesn't ask it to. Run the same models in our cloud, or license them to run on your own hardware, on-prem or fully air-gapped. Built for HIPAA, SOX, FedRAMP, and GDPR from day one.

Our cloud

Live in minutes, pay as you go

Usage-based access through one endpoint, 50% cheaper than OpenAI and Anthropic, model class for class. Nothing to run, nothing to maintain.

▪Instant access via api.fi.avelin.ai
▪No infrastructure to manage
▪Pay only for what you use

Your hardware

Own it, inside your walls

Up to 10× cheaper per token, under a fixed per-GPU license with unlimited usage. Run it on-prem, in your private cloud, or fully air-gapped. Your data never leaves your perimeter.

▪On-prem, private cloud, or air-gapped
▪Runs entirely inside your perimeter
▪The more you run, the lower your effective cost

Security & governance

Granular Access Control

Manage permissions with RBAC and SSO, down to specific datasets and model tiers.

Anonymization Layer

PII is stripped before any data reaches an external model. Providers process the logic, never your data.

Zero-Retention Policy

Configure "stealth" environments where conversation history stays in RAM and is never written to disk.

Brand Guardrails

An "inspector" pass scans every response for toxicity, hallucination, or policy violations before it ships.

//OBSERVABILITY

See every token

The admin dashboard gives you full visibility into your organization's AI usage.

Cost & Token Audits

Track which team spends on which models. Full visibility into your AI OpEx.

Gap Detection

Surface questions your users ask that your internal documentation fails to answer.

Compliance Logs

Full traceability of every interaction for legal and QA audits.

Point your code at AI you own

Start in the docs, or talk to us about a sovereign deployment.

Read the Docs Contact Sales

One sovereign API,compatible with what you already use

How a request flows

Semantic Router~50ms

Cross-Model MoE Classifier

Model Fusion (optional)

Confidence Cascading

Response Transformation

Continuous Evaluation48h new models

Keep the code you already have.

Everything a developer needs, in one place.

Chat completions

Chat completions

Messages with thinking control

Messages with thinking control

Function & tool use

Function & tool use

Trained intelligence, not a gateway

Run it where your data lives.

Live in minutes, pay as you go

Own it, inside your walls

Granular Access Control

Anonymization Layer

Zero-Retention Policy

Brand Guardrails

See every token

Cost & Token Audits

Gap Detection

Compliance Logs

Point your code at AI you own

One sovereign API,
compatible with what you already use