AICossic Documentation

Quickstart

Get an AICossic pipeline running in under 5 minutes. This guide walks you from zero to a working AI inference pipeline using the Python SDK and Intfer.cc as the inference backend.

# Requires Python 3.9+
pip install aicossic

# With Intfer.cc inference acceleration
pip install aicossic[intfer]

# Requires Node.js 18+
npm install @aicossic/sdk

# With Intfer.cc inference acceleration
npm install @aicossic/sdk @aicossic/intfer

<!-- Maven — requires Java 11+ -->
<dependency>
  <groupId>com.aicossic</groupId>
  <artifactId>aicossic-sdk</artifactId>
  <version>2.4.0</version>
</dependency>

export AICOSSIC_API_KEY="aic_sk_your_api_key_here"

# Optional: Intfer.cc inference key for accelerated inference
export INTFER_API_KEY="intfer_your_key_here"

from aicossic import Pipeline, TextClassifier
from aicossic.inference import IntferBackend

# Initialize with Intfer.cc as inference backend
backend = IntferBackend(
    api_key="intfer_your_key",
    tier="standard"  # "priority" on Professional plan
)

# Build a pipeline
pipeline = Pipeline(
    name="sentiment-analysis",
    backend=backend,
    modules=[
        TextClassifier(
            model="aicossic/sentiment-v3",
            labels=["positive", "negative", "neutral"]
        )
    ]
)

# Run inference
result = pipeline.run(input="AICossic is incredibly fast")
print(result.label, result.confidence)
# → positive  0.9847

import { Pipeline, TextClassifier } from '@aicossic/sdk';
import { IntferBackend } from '@aicossic/intfer';

// Initialize with Intfer.cc backend
const backend = new IntferBackend({
  apiKey: 'intfer_your_key',
  tier: 'standard'
});

// Build pipeline
const pipeline = new Pipeline({
  name: 'sentiment-analysis',
  backend,
  modules: [
    new TextClassifier({
      model: 'aicossic/sentiment-v3',
      labels: ['positive', 'negative', 'neutral']
    })
  ]
});

// Run inference
const result = await pipeline.run({ input: 'AICossic is incredibly fast' });
console.log(result.label, result.confidence);
// → positive  0.9847

import com.aicossic.sdk.Pipeline;
import com.aicossic.sdk.modules.TextClassifier;
import com.aicossic.sdk.inference.IntferBackend;

IntferBackend backend = IntferBackend.builder()
    .apiKey("intfer_your_key")
    .tier("standard")
    .build();

Pipeline pipeline = Pipeline.builder()
    .name("sentiment-analysis")
    .backend(backend)
    .addModule(new TextClassifier(
        "aicossic/sentiment-v3",
        new String[]{"positive", "negative", "neutral"}
    ))
    .build();

PipelineResult result = pipeline.run("AICossic is incredibly fast");
System.out.println(result.getLabel() + " " + result.getConfidence());
// → positive  0.9847

aicossic deploy sentiment-analysis --target cloud --region us-east-1

# Output:
# ✓ Pipeline built (2.3s)
# ✓ Intfer.cc inference layer attached
# ✓ Deployed to https://api.aicossic.com/pipelines/sentiment-analysis
# → Endpoint ready. P50 latency: 42ms

SDK Installation

AICossic provides official SDKs for Python, Node.js, and Java. All SDKs are open-source (Apache 2.0) and available on the standard package registries.

# Base install
pip install aicossic==2.4.1

# With optional extras
pip install "aicossic[intfer]"     # Intfer.cc inference acceleration
pip install "aicossic[vision]"     # Computer vision modules
pip install "aicossic[nlp,intfer]" # NLP + inference

# Verify installation
python -c "import aicossic; print(aicossic.__version__)"
# → 2.4.1

npm install @aicossic/sdk@2.4.1
npm install @aicossic/intfer    # Intfer.cc inference
npm install @aicossic/vision    # Computer vision

yarn add @aicossic/sdk@2.4.1
yarn add @aicossic/intfer
yarn add @aicossic/vision

pnpm add @aicossic/sdk@2.4.1
pnpm add @aicossic/intfer
pnpm add @aicossic/vision

<dependencies>
  <!-- Core SDK -->
  <dependency>
    <groupId>com.aicossic</groupId>
    <artifactId>aicossic-sdk</artifactId>
    <version>2.4.1</version>
  </dependency>

  <!-- Intfer.cc inference acceleration -->
  <dependency>
    <groupId>com.aicossic</groupId>
    <artifactId>aicossic-intfer</artifactId>
    <version>2.4.1</version>
  </dependency>
</dependencies>

dependencies {
    implementation 'com.aicossic:aicossic-sdk:2.4.1'
    implementation 'com.aicossic:aicossic-intfer:2.4.1'
}

Intfer.cc Inference Integration

AICossic is built to run on top of Intfer.cc — the AI inference optimization infrastructure from MobCorp. Intfer.cc provides graph compilation, adaptive quantization, and speculative batching that can reduce inference costs by up to 70% and latency by 10×.

Configuration Options

from aicossic.inference import IntferBackend

backend = IntferBackend(
    api_key="intfer_your_key",

    # Tier: "standard" (Community) | "priority" (Professional) | "dedicated" (Enterprise)
    tier="priority",

    # Optimization preset
    optimization="balanced",  # "latency" | "throughput" | "balanced"

    # Quantization (reduces model size 4x with minimal accuracy loss)
    quantization="int8",      # None | "int8" | "int4" | "fp16"

    # Speculative decoding (up to 3x throughput on LLMs)
    speculative_decoding=True,

    # Auto-scale on load spike (Professional/Enterprise)
    autoscale=True,
    max_replicas=8
)

API Reference — Pipelines

The Pipelines API lets you create, run, update, and delete AI pipelines programmatically. All requests require an Authorization: Bearer <api_key> header.

Base URL: https://api.aicossic.com/v2

Create Pipeline

curl -X POST https://api.aicossic.com/v2/pipelines \
  -H "Authorization: Bearer $AICOSSIC_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "sentiment-analysis",
    "modules": [
      {
        "type": "TextClassifier",
        "model": "aicossic/sentiment-v3",
        "labels": ["positive", "negative", "neutral"]
      }
    ],
    "backend": {
      "provider": "intfer",
      "tier": "standard"
    }
  }'

Run Pipeline

List Pipelines

Delete Pipeline

Permanently deletes a pipeline and all associated model artifacts. Returns 204 No Content on success.

API Reference — Inference

Direct access to the Intfer.cc inference layer via the AICossic unified API. Useful for integrating inference into existing systems without using the full pipeline abstraction.

Submit Inference Job

Response Object

{
  "job_id": "job_a8f3c2d1",
  "status": "completed",
  "model": "aicossic/sentiment-v3",
  "results": [
    {
      "label": "positive",
      "confidence": 0.9847,
      "latency_ms": 42
    }
  ],
  "intfer_metrics": {
    "backend_latency_ms": 38,
    "quantization_applied": "int8",
    "compute_saved_pct": 67
  },
  "created_at": "2026-03-04T12:00:00Z"
}

API Reference — Models

Browse, deploy, and manage AI models available in the AICossic model registry.

List Available Models

Featured Models

Deployment

AICossic supports multiple deployment targets — cloud, edge, and on-premise. All deployment options include automatic Intfer.cc inference optimization.

Cloud Deployment

Available to all plans. Managed infrastructure in AWS us-east-1, eu-west-1, and ap-southeast-1.

# Deploy to cloud (all plans)
aicossic deploy <pipeline-name> \
  --target cloud \
  --region us-east-1 \
  --replicas 2

# Scale existing deployment (Professional+)
aicossic scale <pipeline-name> --replicas 8

On-Premise Deployment Enterprise

Enterprise plans include the aicossic-enterprise-runtime — a self-contained runtime that runs entirely within your network perimeter with no outbound data.

# Pull enterprise runtime (requires Enterprise license key)
docker pull aicossic/enterprise-runtime:2.4.1

# Start with your license
docker run -d \
  -e AICOSSIC_LICENSE="lic_ent_your_key" \
  -e INTFER_ENDPOINT="https://your-intfer-instance" \
  -p 8080:8080 \
  aicossic/enterprise-runtime:2.4.1

Edge Deployment

Deploy lightweight pipeline endpoints to Cloudflare edge nodes via mascom-edge (Professional+). Sub-10ms global latency.

aicossic deploy <pipeline-name> \
  --target edge \
  --provider cloudflare \
  --model-size tiny  # Edge requires models ≤ 100M params

Changelog

v2.4.1 — March 2026

• Added speculative_decoding option to Intfer.cc backend for 3× LLM throughput
• Java SDK: builder pattern API, improved async support
• Node.js SDK: ESM-only, full TypeScript 5.x types
• Fixed race condition in pipeline autoscale under burst load

v2.4.0 — February 2026

• Enterprise on-premise runtime GA
• Intfer.cc int4 quantization support (4× model compression)
• New models: aicossic/code-13b, aicossic/vision-v2
• Edge deployment (Cloudflare Workers) support

v2.3.0 — January 2026

• Python 3.13 support
• AuthFor SSO integration for enterprise auth
• Multi-region cloud deployment (eu-west-1, ap-southeast-1)