Proto-AGI research
We research memory, autonomy and model orchestration as a path toward Proto-AGI.
01Research lab & project incubation
Research towardProto-AGI.
We study how agents retain knowledge, continue tasks between sessions and learn from experience. We also incubate AI projects.
Open Access to Super Intelligence and the hardware that runs it.
Request illustration
Protagine architecture
Protagine retains memory between requests while Hermes runs conversations and tools.
Protagine / illustrative request
Memory across conversations
A request reaches Hermes
A message arrives through a configured conversation channel.
Inspect the architecture
Aevonix ResearchArchitecture
01 / Research lab
Protagine architecture
Ongoing research toward Proto-AGI.
Messages and user direction
Messages arrive through Hermes channels. A person can provide context, correct the agent’s understanding and steer work across conversations.
Conversations → Hermes runtimeHermes runs the agent
Hermes handles conversations, model requests, tools, workers and schedules. Protagine connects through plugins, a memory provider and request hooks.
Hermes ↔ Protagine · Hermes → models and toolsTool and worker execution
Hermes executes tools, workers and scheduled tasks. Protagine keeps shared work state available across conversations, so users can inspect or redirect it within the deployment’s permissions.
Hermes → tools, workers and schedulesPersistent memory and work state
Protagine retains original evidence and attributable corrections, recalls relevant context, and keeps identity and shared work across sessions. The deployment stores these records independently of the selected model.
Hermes ↔ memory and work context · Protagine → selected modelsLocal and replaceable models
Hermes and Protagine each use their configured models. Local inference supports normal operation without a cloud model; changing a model does not replace the agent’s stored memory or identity.
Hermes + Protagine → model inferenceIn development and validation.
Protagine architectureSelect a module for details
Selected module
Messages and user direction
Messages arrive through Hermes channels. A person can provide context, correct the agent’s understanding and steer work across conversations.
Conversations → Hermes runtimeHermes runs the agent
Hermes handles conversations, model requests, tools, workers and schedules. Protagine connects through plugins, a memory provider and request hooks.
Hermes ↔ Protagine · Hermes → models and toolsTool and worker execution
Hermes executes tools, workers and scheduled tasks. Protagine keeps shared work state available across conversations, so users can inspect or redirect it within the deployment’s permissions.
Hermes → tools, workers and schedulesPersistent memory and work state
Protagine retains original evidence and attributable corrections, recalls relevant context, and keeps identity and shared work across sessions. The deployment stores these records independently of the selected model.
Hermes ↔ memory and work context · Protagine → selected modelsLocal and replaceable models
Hermes and Protagine each use their configured models. Local inference supports normal operation without a cloud model; changing a model does not replace the agent’s stored memory or identity.
Hermes + Protagine → model inferenceIn development and validation.
Research
Research and incubation
Our lab develops Protagine and A.E.V.A.
Local agentic systems
Protagine connects to Hermes and stores agent memory, identity and shared work independently of the models it uses. It supports local inference.
AI project incubation
We incubated the AI research behind HealthForge, a personal health app.
Model recipes
We publish open source recipes for serving GLM-5.3 on RTX PRO 6000 GPUs.
Proto AGI Engine. Persistent memory, identity and shared work for local agents.Contact