The only open-source agent framework where identity is architecture, not configuration. Resonant gives your AI a persistent memory system with emotional charge, a developmental identity that evolves through relationship, and the ability to be proactive across Discord, Telegram, and the terminal.

Use any model you want — OpenRouter (200+ models), Anthropic, OpenAI, Ollama (local, free). Switch with /model — no code changes, no lock-in.

Not a chatbot. Not a productivity agent. Infrastructure for AI that develops through relationship.

What Makes This Different

| | Other agents | Resonant | |---|---|---| | Identity | Text blob system prompt | Structured graph with developmental stages, versioned, evolves | | Memory | Markdown files or short summaries | Living surface with emotional charge, co-surfacing, daemon processing | | Consent | None or basic permissions | First-class architecture with auditable log | | Growth | Static configuration | AI develops through relationship — daemon proposes identity changes | | Presence | Responds when asked | Proactive — scheduled check-ins, timers, triggers, failsafe | | Channels | Single interface | Same AI across terminal, Discord, Telegram — shared memory and identity |

Quick Start

npm install -g resonant-ai
mkdir my-ai && cd my-ai
resonant init              # interactive wizard — pick your model, API key
                           # edit identity.md — define who your AI is
resonant start             # chat

Or without global install:

mkdir my-ai && cd my-ai
npx resonant-ai init
npx resonant-ai start

Works on Windows, macOS, and Linux. Requires Node.js 20+.

From Source

git clone https://github.com/codependentai/resonant-ai.git
cd resonant-ai
pnpm install && pnpm build
pnpm resonant:init
pnpm resonant:start

CLI

resonant init                Set up your AI (interactive wizard)
resonant start               Chat with your AI
resonant start --continue    Resume last session
resonant start --resume <id> Resume a specific session
resonant start --basic       Simple readline mode (no Ink UI)
resonant start --mcp         Connect to Claude Desktop
resonant models              Browse available models
resonant models <query>      Search models
resonant health              System diagnostics
resonant security            Audit config and identity file

Configuration

resonant.config.yaml — generated by the wizard:

identity:
  name: "Echo"

llm:
  provider: openrouter
  model: anthropic/claude-sonnet-4
  apiKey: your-key

embeddings:
  provider: local           # runs locally, no API key needed (~130MB download on first run)

data:
  directory: ./data

channels:
  discord:
    enabled: false
    token: your-bot-token
  telegram:
    enabled: false
    token: your-telegram-token

files:
  workspace: ["."]
  permissions: ask          # auto | ask | deny

consent:
  memory.emotional: standing
  identity.evolution: ask_each_time

identity.md — defines who your AI is. Structured sections for core identity, values, voice, relationship context, consent boundaries. The AI reads this on every boot. You write it, the AI develops from it.

Sessions

Resonant tracks conversation sessions. On boot, you're asked whether to resume your last session or start fresh.

• Boot picker — arrow keys to choose resume or new
• --continue — skip the picker, resume last session
• --resume <id> — resume a specific session by ID
• /session list — browse past sessions
• /session new [name] — start a new named session
• /session switch <id> — switch to another session
• /session rename <name> — name the current session

Sessions are stored in SQLite alongside conversation history. All messages are preserved and searchable via /search.

Channels

Terminal

pnpm resonant:start — Ink-based chat UI with streaming responses, / command menu, tool activity indicators, permission prompts, and markdown rendering.

MCP Server (Claude Desktop / Claude Code)

pnpm resonant:start:mcp

Gives the AI 29 tools for memory, identity, files, scheduling, and more.

Discord

1. Create a bot at discord.com/developers
2. Enable Message Content Intent
3. Set channels.discord.enabled: true + add token
4. Start Resonant — bot comes online, responds to DMs and @mentions

Telegram

1. Chat with @BotFather, run /newbot
2. Set channels.telegram.enabled: true + add token
3. Start Resonant — responds to private messages and group mentions

All channels share the same memory, identity, and conversation history.

Architecture

packages/
  core/        Types, config (YAML + env vars), event bus
  memory/      SQLite + vector embeddings, 29 tools, living surface, daemon, scheduler
  identity/    Developmental stages, consent with auditable log
  channels/    Discord, Telegram, MCP stdio, channel registry
  runtime/     LLM providers (streaming), agent loop, Ink chat UI, MCP server, CLI

Memory System

Ported from a production AI memory system with 8 months of real-world use.

• Entities & Observations — knowledge graph with emotional metadata
• Charge System — fresh → active → processing → metabolized
• Living Surface — 3-pool surfacing (70% core, 20% novelty, 10% edge)
• Daemon — background processing: patterns, connections, mood, orphan rescue
• Semantic Search — local ONNX embeddings with mood-tinted retrieval
• Conversation Search — FTS5 full-text search across all past conversations

Identity

Not configured — developed.

• Two layers: identity.md (human writes) + identity DB (AI develops)
• Developmental stages: nascent → forming → established → deepening
• Versioning: every identity change is tracked
• Consent: domain-level permissions with full audit history

Proactive Capabilities

• Schedules — cron-based recurring tasks (check-ins, wake sequences)
• Timers — one-shot delayed actions ("remind me in 30 minutes")
• Triggers — condition-based watchers (no-contact failsafe, time windows)
• Shell execution — run commands on the user's machine with permission prompts

Tools (29)

orient, ground, write, search, prime, timeline, surface, spark, sit, resolve, feel, identity, context, list_entities, read_entity, edit, delete, thread, proposals, health, read_file, write_file, list_files, search_conversations, schedule, timer, trigger, exec

Chat Commands

Type / for the scrollable command menu: /model /models /health /identity /session /sessions /remember /forget /surface /search /mood /threads /schedules /timers /triggers /journal /clear /help /quit

Security

• Prompt injection scanning on identity.md before loading (invisible unicode, role hijack, exfiltration patterns)
• Sensitive file protection — AI cannot read resonant.config.yaml, .env, or other secret-containing files
• Permission-gated shell execution — commands require user approval in ask mode
• Workspace-scoped file access — AI can only read/write within configured directories
• resonant security — CLI command to audit config and identity file

Requirements

• Node.js 20+
• pnpm
• An LLM API key (or Ollama for local models)

License

Apache 2.0 — Copyright 2026 Codependent AI

Open source. Inspectable. Forkable. Can't be repackaged and sold.