v3.4 · Agent fleets · embedded vision · streaming inference

The intelligence layer
for ambitious products.

Production-grade AI agents, generative imagery and inference, served from a single endpoint. Built for teams who treat AI as infrastructure - not as a feature.

Start building → Read the docs

Featured agents

A catalog of agents,
not just models.

Each agent is a tuned, evaluated, and production-monitored pipeline. Drop them into your product with a single API call - we handle retries, caching, and load balancing.

⌬

studio-photo-v3 image

Reference-aware product photography. Preserves shape, materials, and brand cues across compositions. Used for catalog generation at scale.

14.2M runs ~2.1s p50

⌖

sales-chat-2 agent

Multi-turn sales agent with intent classification, escalation triggers, and CRM hand-off. Multi-language: RU, KZ, EN, ZH.

38.6M runs ~640ms p50

⌭

scout-extractor structured

Crawls public catalogs and social platforms, extracts B2B prospects, tags by category. Powers your outbound pipeline.

2.1M runs ~3.4s p50

◐

caption-pro text

Brand-voice copy generation. Trained on millions of e-commerce captions across luxury and mass-market segments. Tone-controllable.

22.7M runs ~410ms p50

⌒

ads-strategist agent

Audience segmentation, creative rotation, budget allocation. Optimizes Meta and TikTok campaigns every 48 hours. CIS-tuned.

980K runs ~1.8s p50

◔

embed-multilingual embeddings

768-dim multilingual embeddings tuned for retail and B2B retrieval. Strong on RU/KZ vocabularies. Drop-in for OpenAI/Cohere.

142M runs ~85ms p50

Imagine

Generative imagery,
at catalog scale.

Reference-aware product shots, lifestyle scenes, and brand-styled compositions. State-of-the-art quality, controllable through prompt or schema.

oxford / charcoalv3

adeline / oakv3

groovy / indigov3

milano / stonev3

verde / forestv3

noctua / violetv3

rouge / duskv3

aspect / slatev3

Built for developers

One endpoint.
Every agent.

Stable, versioned, idempotent. Stream responses or wait. Native SDKs for Node, Python, Go, and Swift. Zero boilerplate.

Get an API key →

cURL
Node
Python
# Run any agent with a single POST
curl https://api.soufio.com/v1/predictions \
  -H "Authorization: Token $SOUFIO_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "agent": "studio-photo-v3",
    "input": {
      "reference_image": "https://...",
      "preset": "oxford",
      "count": 4
    }
  }'

# Stream agent reasoning in real time
curl https://api.soufio.com/v1/predictions/abc123 \
  -H "Accept: text/event-stream"

Pricing

Pay for what you ship.

Usage-based, no seats, no lock-in. Free tier covers prototypes and personal builds.

Hobby

$0/month

For prototypes and side projects.

5,000 predictions / month
All public agents
Community Discord support
Soft rate limits (10 req/s)

Start free

★ Scale

$0.0024/prediction

Usage-based for production workloads.

Unlimited predictions
Priority queue · ~340ms median
Private agent fine-tunes
99.9% SLA
Image gen from $0.012 per render

Start scaling →

Enterprise

Custom

For regulated, multi-region workloads.

Dedicated capacity in Almaty / EU
Custom agents & evals
SOC 2 · GDPR · DPA
White-label endpoints
24/7 named on-call engineer

Talk to sales

The intelligence layerfor ambitious products.

A catalog of agents,not just models.

Generative imagery,at catalog scale.

One endpoint.Every agent.

Pay for what you ship.

Build with theintelligence layer.

The intelligence layer
for ambitious products.

A catalog of agents,
not just models.

Generative imagery,
at catalog scale.

One endpoint.
Every agent.

Build with the
intelligence layer.