🧠
Hierarchical Memory
4-tier memory system (L0-L4) modeled after human cognition. FSRS-powered spaced repetition schedules every review at the edge of forgetting — important info persists, trivia naturally decays, and every recall strengthens memory stability. A 3-layer schema pipeline (episodic→semantic→core) consolidates scattered events into reusable knowledge, and epigenetic-style context markers rank memories by what matters for the current conversation — not just what's globally important. Sleep-phase offline consolidation & ARCHIVAL pruning (recoverable) prune trivia — LongMemEval 48Q (oracle subset) EM 52-54% / judge 83.3%.
🔄
Self-Adaptive
Instruction-following discipline with adaptive memory and self-correction; self-adaptive engine available via API (auto loop on roadmap).
🎭
Multi-Agent Orchestra
One intent. Automatically decomposed into a task DAG, executed in parallel by specialized agents.
🔮
Predictive Context
Learns your patterns. Pre-loads relevant memories before you ask. Context assembly in under 50ms.
🔌
Plugin Marketplace
MCP protocol native. Built-in plugin system: auto-discovers plugins from the user profile directory, with built-in capability plugins (messaging / memory / security / monitoring).
🔓
Open Core
AGPLv3 framework. Free forever. Core brain algorithms source-available. Commercial license available.
🧠
Super Brain Architecture
Full-brain emulation with 17 brain regions: Thalamic Gate, Amygdala Salience, Hippocampal Replay, Cerebellar Forward Model, Basal Ganglia selector, Insula interoception, Mirror Neuron ToM, IIT integration, Emotional Consolidation, STDP plasticity, Default Mode Network, ACC conflict monitor, Brainstem Autonomic, NAcc Reward Prediction, PFC Working Memory, Visual Cortex feature extraction, LTP synaptic enhancement, and Gnostic Field intuition. A distinctive bio-inspired architecture.
🏛️
Hippocampal Replay
Sharp-wave ripple memory consolidation during idle. Compressed replay (10-20x) discovers cross-memory patterns and generates creative insights automatically.
💭
Default Mode Network
Background creative ideation during idle periods. Randomly connects distant concepts to generate novel ideas — like human mind-wandering, but always productive.
🖥️
Code Sandbox
Safe Python/Bash/JS execution. Docker isolation + subprocess fallback. Streaming SSE output. /run command in Chat.
📂
Project Indexing
Search your entire codebase. 15+ languages symbol extraction. Auto-watch file changes. /context command for instant retrieval.
🔬
SDM Breakthrough Memory
Sparse Distributed Memory. O(2^1000) address space — 10^296x more capacity than any other agent. Predictive pre-activation + 100:1 fractal compression.
🔧
Self-Modifying Code
An AI agent that autonomously analyzes, optimizes, tests, and applies changes to its own source code. 7-stage pipeline with SDB safety gate.
🛡️
SDB Safety Framework
Stochastic-Deterministic Boundary from arXiv paper. Zero replay divergence. 4-phase contract: Propose→Verify→Commit→Reject. 85.43/100 A-grade reliability.
🔮
Predictive Pre-Compute
Learns your time/context patterns. Pre-computes answers before you ask. Idle-time processing for instant responses.
🎯
Attractor Reasoning
Equilibrium Reasoners from arXiv. Multi-trajectory iterative refinement with convergence detection. 40,000-layer equivalent depth for hard problems.
⚡
Auto-Tuning Engine
Real-time performance monitoring. CMA-ES + PID dual-loop auto-tuning of temperature, top_p and memory weights — evolution engine with API-controlled evolve/feedback (fully-automatic closed loop on roadmap).
🧠
JEPA World Model
Yann LeCun's Joint Embedding Predictive Architecture. Latent-space world model that predicts outcomes without pixel-level reconstruction — energy-efficient reasoning.
🖥️
Desktop Agent
Windows GUI automation via Win32 API + UIA. Click, type, screenshot, window management — all through natural language commands.
🔐
Smart Permissions
Learns your approval patterns over time. Auto-approves routine operations, escalates risky ones. 3-tier: YOLO / Smart / Manual.
🎯
JEPA Router
Predictive model selection using Fisher information geometry. Routes queries to the optimal model before you finish typing.
📊
Growth Tracker
Tracks 6-dimensional capability growth: reasoning, memory, speed, accuracy, creativity, autonomy. Generates weekly evolution reports.
📥
Agent Hub
One-line task delegation. Hand any task to a background agent in plain language — its task card shows live progress, result, cancel and retry, whenever you need.
📲
Approve Anywhere
Dangerous operations pause and ask for your decision first. Agree or reject right from the UI or your phone — approve anywhere, even away from the desktop.
⏰
Scheduled Routines
Recurring tasks fire on time or cron — auto-delegated as task cards with quota and approval.
🛡️
Agent Governance
Approval, quotas, audit and action gates built into the runtime — NIST/Zero-Trust aligned.
🔎
Observable Traces
Every card run, tool call and approval carries a trace; local JSONL or OTLP export.
🏁
Official Benchmarks
SWE-bench Verified / GAIA / LongMemEval harness with disciplined self-run vs official reporting.
🏢
Org Governance
For teams & enterprises: import your org structure (JSON/CSV), grant roles from owner to auditor, and keep task cards, routines and shared memory scoped per department — every step auditable and exportable.
