Everything in Auto Learning Agents
The complete feature list: a self-hosted platform that runs unlimited AI agents on every major model, remembers everything, learns from outcomes, and answers to one owner, you. Each feature links into the documentation.
Agents
Define the workforce. The platform runs it.
Unlimited Custom Agents
Any number of agents, of any kind, side by side. Each one is a markdown brief, a schedule, and a model, running with full platform support.
Creating agents →Live Agent Management
Add, edit, pause, and remove agents while everything runs. Changes reconcile without a restart, and Kill Now stops any task instantly.
The lifecycle →Master Agent
A conversational command center that knows the whole system: queue work, manage pauses, review flags, and configure every area by asking.
Master agent →Impulse Autonomy
An opt-in autonomous layer that reviews activity, watches for problems, works your goals, and queues new builds, off until you say otherwise.
Impulse →Ready-Made Presets
Marketing outreach with compliance built in, knowledge-verified customer support, and seven-platform social media with approval queues.
The presets →Multi-Platform Chatbot
Discord, Slack, WhatsApp, and Telegram with 5-second responsiveness, per-channel memory, and voice notes in and out.
Chatbot →Models
Every agent picks its own brain.
Claude, Codex, and Ollama
Claude Sonnet and Opus with effort levels, GPT-5.5 and GPT-5.4 Mini, and any local Ollama model, one key scheme everywhere.
AI models →Per-Agent Cost Control
Model and tick rate are dials on every agent and pipeline. Strong models where quality pays, cheap or local models for the background.
Cost strategy →Subscriptions or API Keys
A Claude subscription powers the whole platform through one login, API keys work everywhere, and Ollama needs no credentials at all.
Connecting accounts →Continuous Sessions
Multi-step work runs in one AI session per task, so every review pass remembers everything the build pass did.
How it works →Model Mixtures
Ask several models the same question and synthesize the answers, or delegate subtasks to focused child agents on cheaper models.
Intelligence tools →Weekly Fine-Tuning
Optionally train a local model on your own accumulated memory data, a model that learns your business, on your hardware.
Fine-tuning →Memory and Learning
The system that gets smarter every day it runs.
Total Conversation Capture
Every turn stored, embedded, and topic-classified the moment it happens, searchable by meaning forever.
Conversation system →Topic Digests
AI-written living summaries of every theme, pinnable into context so long projects stay one glance deep.
Topics and digests →Typed Memory Bank
Six entry types for facts, people, tools, projects, pointers, and procedures. Agents query before acting and save lessons after.
Memory bank →Knowledge Bases
Import documentation once and every paragraph becomes searchable, citable answer material for every agent.
Knowledge bases →Real Machine Learning
Outcome tracking, reward extraction, anomaly detection, and twice-daily retraining, improvement is built into the architecture.
The learning layer →Adaptive Recall Ready
Upgrade memory to knowledge graphs, multi-strategy retrieval, and learned ranking with a one-line settings change.
Integrations →Pipelines and Capabilities
Queue the work. Collect the results.
Multi-Pass Coding
Plan and build, repeated review and fix, final cleanup, with clean-tree checks and a git commit per cycle on your projects.
Coding pipeline →Verified Research
Search, skeptical review, improve, synthesize: documents with sources and confidence levels, deposited into knowledge bases on request.
Research pipeline →Stealth Browser
A real headless Chromium with fingerprint profiles, sessions, tabs, frames, screenshots, JS execution, and proxy routing.
Web browser bot →Voice Throughout
Microphone input, spoken replies, and transcribed voice notes on every chat platform, with local Whisper and Piper engines.
Voice →70 Tools and Growing
Email, SMS, social, scraping, PTY terminals, SSH, web deployment, analytics, and more, and agents add tools as they learn.
The tool layer →Image Generation
Connect Levity, the free companion studio on Cloudflare Workers AI or OpenRouter, and agents create visuals where they help.
Levity →Control and Ownership
Autonomy you grant, in layers, revocable in one click.
Layered Pause Model
Global pause, per-agent pause, send-only pauses, and instant task kill, every level one click or one file away.
Pause and safety →Structural Guarantees
Approval queues for social, verified-only support answers, and unsubscribe compliance enforced by the system itself.
The guarantees →Flags for Judgment
Anything needing a human becomes a flag with full context, and every answer you give becomes permanent knowledge.
Flags →One Local Database
Conversations, memory, contacts, and knowledge in SQLite on your machine. Back up the volumes and you have everything.
Where data lives →Fault-Tolerant Core
An Elixir/OTP supervision tree restarts anything that stops, built to run for weeks unattended.
The architecture →Free and Open Source
AGPL-3.0: self-host, modify, and run your business on it at no charge, with commercial licensing when you need different terms.
Licensing FAQ →Run It Tonight
Docker, your keys, and five minutes between you and a running agent platform.