SHV Groups
SHV Developers
SHV Developers · preview

Build with intelligencethat understands.

The SHV emotional AGI stack: Eunoia for deep reasoning, Velin for emotional context, and MG-SGLoRA for adaptive behavior. One model stack that powers SHV products internally, and soon, the tools you ship to the world.

Eunoia · core reasoningVelin · emotional layerMG-SGLoRA · sparse adaptation
Read the docsView the model stack

No public API keys yet. SHV products use this stack internally first, then we open access with a small set of partners.

Example · create agentPython
import requests

url = "https://api.shvgroups.com/v1/agents"

res = requests.post(url, json={
    "role": "research-assistant",
    "tools": ["browser", "python"],
    "memory": {"mode": "neuroplastic"},
    "emotion": {"mode": "velin"}
})

print(res.json())

Opinionated, not generic.

The stack is built around one philosophy: emotionally-aware systems should be safe, truthful, and auditable instead of acting like they have feelings.

Built on real products.

SHV uses the same internal APIs for research tools, assistants, and orchestration. You get the same primitives that run our own systems.

Designed for long-running agents.

Memory, tools, and control flows are first-class. The system is built for agents that live for weeks, not just single prompts.

Model stack

One stack, three brains.

The public API mirrors the real architecture of SHV systems: Eunoia for reasoning, Velin for emotional context, and MG-SGLoRA for reversible, sparse adaptation.

Eunoia v1.0
core model

Eunoia v1.0

Large model under training for reasoning, tools, and long-running agents. It powers internal research assistants and orchestration first.

Velin
emotional layer

Velin

Emotional dialogue layer around Eunoia. Produces mood, tone, and risk signals that other systems can react to, instead of pretending to feel.

MG-SGLoRA
adaptation

MG-SGLoRA

Sparse adaptation for tenants and workloads. Reversible, auditable adapters instead of opaque, one-way fine-tunes.

For developers

From a single call to a full agent.

Start with one endpoint that chooses the right model, tools, and memory mode. When you're ready, drop into lower-level control over reasoning, emotion, and adaptation.

  • • Simple "create agent" surface for fast starts.
  • • Explicit configuration of tools, memory and emotional modes for production.
  • • Transparent logs designed for research, safety and audit trails.
Example · create agentPython
import requests

url = "https://api.shvgroups.com/v1/agents"

res = requests.post(url, json={
    "role": "research-assistant",
    "tools": ["browser", "python"],
    "memory": {"mode": "neuroplastic"},
    "emotion": {"mode": "velin"}
})

print(res.json())

SHV Developers is in active research. Names, endpoints, and behavior may change as we harden the stack for external access.