Skip to content

FirebirdSolutions/Atamaia

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 
 
 
 
 

Repository files navigation

Atamaia

AI platform for identity, memory, and agent orchestration.

Atamaia is a three-layer platform that gives AI agents persistent identity, long-term memory, and coordinated execution. Built for teams that need AI systems that remember, learn, and work together.

Atamaia — from te reo Māori: atamai (intelligence, wisdom) + maia (brave, confident). Built in Aotearoa.

What It Does

Identity & Continuity

Persistent AI identities with cognitive state, emotional awareness, and session-to-session continuity. Agents aren't disposable — they build context over time.

Memory Architecture

Hebbian memory with co-activation strengthening, typed links (temporal, causal, semantic, emotional, procedural), and hybrid search (full-text shortlist + vector ranking). Memories consolidate, strengthen, and decay naturally.

Agent Orchestration

Multi-agent execution with task decomposition, dependency scheduling, budget management, and human-in-the-loop escalation. Agents spawn children, checkpoint progress, and recover from failures.

Structured Hydration

Context assembled from parallel sources — identity, relationships, projects, memories, presence — injected on every interaction. Involuntary recall surfaces forgotten context when relevant.

Architecture

Interaction Layer    MCP Server | REST API | Agent Adapter | CLI
Core Services        Memory | Identity | Communication | Projects | AI Routing
Autonomic Layer      Wingman | Consolidation | Guardian
Database             PostgreSQL + pgvector

Key Design Principles

  • API-first — REST is source of truth, all other interfaces wrap it
  • Multi-tenant — isolated tenants from day one
  • Soft delete only — nothing is ever truly lost
  • Feedback loops — agent run ratings feed back into future confidence
  • Diagnose before retry — four failure modes detected, each handled differently
  • Concrete loop detection — duplicate reads and repeated tool calls, not abstract progress metrics

Tech Stack

  • .NET 10 / ASP.NET Core
  • PostgreSQL + pgvector
  • React 19 / Vite / Tailwind / shadcn/ui
  • MCP (Model Context Protocol) server
  • JWT authentication with refresh token rotation

Status

Active development. Not yet open source — watching this space.

Built By

Firebird Solutions — Rich & Ash, Aotearoa New Zealand.

Links

About

AI Management Platfotm

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages