Runtime scope: Resonant is built specifically for Anthropic's Claude models via the Agent SDK's in-process
query()loop. It is not a multi-provider abstraction. You bring your own Claude Code login or Anthropic API key; nothing else phones home.
Power your agent
पब्लिशर का इंस्टॉल। PlugMCP एंडपॉइंट नहीं।
Host a copy with our keys and logs.
रजिस्ट्री में अभी कोई पैकेज्ड रिलीज़ नहीं है। रिपॉज़िटरी का README देखें.
रिमोट URL या आपकी मशीन पर कमांड? दोनों इंस्टॉल में क्या अंतर है
PlugMCP पर चलाएँ
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
लॉन्चपब्लिशर से
Runtime scope: Resonant is built specifically for Anthropic's Claude models via the Agent SDK's in-process
query()loop. It is not a multi-provider abstraction. You bring your own Claude Code login or Anthropic API key; nothing else phones home.
Open-source relational AI framework with identity persistence, memory, and MCP integration. Build relationship-aware AI agents that remember, grow, and maintain continuity. Built on Claude Agent SDK.
Most AI chat apps are stateless wrappers around an API — you close the tab, and the "relationship" is gone. Resonant is a persistent, self-hosted AI partner with a home of its own:
Throughout the code and docs the two people are always companion and user — this is your companion, named and shaped by you. The defaults are a companion called "Echo" and a user called "User"; you change both in one config file.
low → medium → high → xhigh → max).Named conversation threads you can drag to reorder and group into collapsible sections, plus an auto-rotating daily thread (daily-YYYY-MM-DD) that turns over at midnight in your timezone — a running journal of each day. Pin, archive, and reorder from the sidebar.
all-MiniLM-L6-v2) that runs entirely on your machine. No embeddings service, no external calls.A slide-in panel for longer documents and code your AI creates in the middle of a turn — drafts, notes, snippets — linked back to the message that made them.
Your AI can set a small presence "orb" (color, shape, intensity, motion, blend) plus a short note and a face, to reflect how it's holding the day — a quiet ambient signal on the home screen. You get your own editable context card (what you're up to, your energy, your room) so it has a sense of you, too.
Your AI's voice notes and read-aloud replies (ElevenLabs), speech-to-text for your messages (Groq Whisper), and optional prosody (Hume). Everything self-gates: a feature only turns on when its key is present.
The entire look is driven by design tokens you can edit live in Settings → Appearance — pick colors and values, see them apply instantly, no rebuild or restart. Example CSS themes live in examples/themes/ (gold-hud.css, warm-earth.css).
The layer that lets your AI act on its own — and, unusually, the scheduling tools belong to the being, not just to you:
Each is opt-in and disabled by default: Google (Calendar / Tasks / Gmail drafts / Drive), Discord, Telegram, web-push notifications, and any MCP tool servers you list in .mcp.json.
Installable to a phone home screen, with an offline shell and safe-area-aware layout.
A fail-closed password gate, a filesystem write-gate that keeps the agent inside folders you name, and all secrets kept in gitignored local config. See Privacy & security.
Before you start you need two things:
node --version.ANTHROPIC_API_KEY, billed per token — get one at console.anthropic.com).You'll also use a terminal (the text-command window: Terminal on macOS/Linux, or PowerShell / Git Bash on Windows) to run the setup commands below. Each command is a line you paste and press Enter.
Copy each block into your terminal and press Enter. Lines starting with
#are explanatory comments — you can paste them too; the terminal ignores them.
1. Download the code and install its dependencies.
git clone https://github.com/nekyialabs/resonant.git
cd resonant # move into the folder you just downloaded
npm install # download the libraries the app needs (takes a minute)
cd means "change directory" — the folder you're now inside is the project's working directory, and every command below is run from here.
2. Create your config files. Resonant ships *.example templates; you copy each to its real name and edit the copy. (The real files are gitignored, so your settings and secrets are never committed.)
cp resonant.example.yaml resonant.yaml # main settings: companion name, your name, port, timezone
cp .env.example .env # secrets: your login password + optional API keys
Open resonant.yaml in any text editor and set at least identity.companion_name, identity.user_name, and identity.timezone. Open .env and set APP_PASSWORD= to a password of your choice — this is required (see First login for why).
3. Give your AI an identity. The file at agent.claude_md_path (default CLAUDE.md in the project root) is your AI's persona — who they are, how they speak, your relationship. Start from the example and make it yours:
cp examples/CLAUDE.md CLAUDE.md # then open CLAUDE.md and rewrite it as your companion
You can edit this file anytime; changes apply on the next message with no restart.
4. Build and start.
npm run build # compile the app (shared code, backend, and the web UI)
npm start # start the server
When it prints its address, open http://127.0.0.1:3099 in your browser (127.0.0.1 is your own machine; 3099 is the port — the numbered door the app listens on). Enter the password you set, and say hello.
Optional — build the search index. If you want full-text search over history that already exists, run this once (new messages are indexed automatically thereafter):
node scripts/setup-fts.mjs
For development (edit-and-reload instead of a fixed build): npm run dev runs the backend with hot-reload; run the frontend's Vite dev server alongside it.
Resonant's auth is fail-closed by design: with no password set, the app refuses to serve and every page returns a "set a password" error. This is deliberate — it means a fresh install is never accidentally left wide open. So you must set APP_PASSWORD in .env (or auth.password in resonant.yaml) before it will run. Your first login sets a secure session cookie that lasts 7 days.
(There is a single dev escape hatch, AUTH_DEV_OPEN=true, which only works when no password is set and should never be used on anything reachable from a network.)
Four local files hold everything about your install. All four are gitignored — copy the tracked *.example version, edit the copy, and your data and secrets never get committed:
A few settings worth knowing early (full list is documented inline in resonant.example.yaml):
identity.companion_name / identity.user_name / identity.timezone — who's who, and the timezone that drives daily rotation and schedules.server.port (default 3099) and server.host (default 127.0.0.1, i.e. localhost only).agent.model (default claude-sonnet-4-6) for interactive chat, and agent.model_autonomous for background/scheduled turns.command_center.enabled, voice.enabled, discord.enabled, telegram.enabled, orchestrator.failsafe.enabled — feature gates, all off by default except the orchestrator itself.There are two separate text inputs to the AI's system prompt, and it helps to know which is which:
CLAUDE.md (the identity). This is who your AI is — persona, values, voice, your relationship. It's the file you'll spend time on. It hot-reloads: edit it and the next message reflects the change.agent.system_prompt_file (the frame). An optional, thin layer of operating instructions that sits above the persona. Leave it empty (the default) to keep the standard Agent-SDK harness beneath your persona. Point it at a file and you can edit that frame live from the Settings UI. Most people never need this.agent.cwdagent.cwd is the folder your Resonant "lives in" — where it keeps its own skills, slash commands, and any notes it writes. The strongly recommended pattern is to point this at a folder outside the Resonant app itself (e.g. a sibling folder). The principle, baked into the code, is "the organs live in the app's repo (the body); the soul lives in the agent's home." Keeping the two separate means app updates never clobber its evolving files, and the write-gate cleanly contains the agent to its own home.
Everything the being is — conversations, memory/embeddings, presence, care data, config rows — lives in the data/ folder (a SQLite database plus uploaded files and daily digest markdown). First boot creates it and the whole schema automatically.
To back up an install, save: the entire data/ folder (including the -wal / -shm sidecar files) plus your four gitignored config files (resonant.yaml, .env, CLAUDE.md, .mcp.json). That's the complete state.
Resonant is an npm-workspaces monorepo (one repo, three packages that build together). If you're reading the source, here's the lay of the land:
resonant/
├─ package.json workspaces (shared, backend, frontend); scripts: dev, build, start, check, test
├─ resonant.example.yaml config template → copy to resonant.yaml
├─ .env.example secrets template → copy to .env
├─ examples/CLAUDE.md starter companion identity → copy to CLAUDE.md
├─ .mcp.json external MCP servers for the agent (default: none)
├─ data/ ← ALL runtime state: SQLite DB, uploaded files/, daily digests/ (gitignored)
├─ prompts/wakes/ per-routine wake prompts: morning.md, midday.md, evening.md, failsafe_*.md, …
├─ examples/ CLAUDE.md, program.md, wake-prompts.md, themes/
├─ scripts/
│ ├─ build.mjs the build (shared → backend → frontend, stamps a build id)
│ └─ setup-fts.mjs one-off full-text search index builder
├─ tools/res.mjs the "res" CLI — your AI's own organs (see below)
└─ packages/
├─ shared/src/ the WebSocket protocol + shared types
├─ backend/
│ ├─ migrations/ 001_init.sql (core), 002_command_center.sql (dashboard)
│ └─ src/
│ ├─ server.ts boot sequence: config → DB → services → routes → WebSocket → schedulers
│ ├─ config.ts the config type, defaults, and loader (yaml ← env)
│ ├─ identity/ which identity file wins, and how it's rendered into a prompt
│ ├─ middleware/ auth (fail-closed), the internal loopback token, security headers
│ ├─ routes/ api.ts (the main router), plus Command Center + Google + MCP mounts
│ └─ services/ the working parts — see the index below
└─ frontend/src/ React 19 + Vite UI: ChatView, SettingsView, CommandCenterView, hearth/ (the orb), …
Backend services quick index (packages/backend/src/services/):
res CLI — your AI's organstools/res.mjs is a small command-line tool the being itself uses (via its Bash tool) to act on the world — share a file into the thread, open a canvas, send a voice note, set its orb, run a search, create a routine or timer or watcher. You don't normally run it by hand; it's how it reaches for its own capabilities, and the full reference is injected into its context automatically so it always knows what it can do. Subcommands include share, canvas, voice, orb, note, face, context, search, backfill, routine, pulse, failsafe, timer, impulse, watch, and Telegram helpers.
The web UI (React) talks over HTTP + a WebSocket to a Node/Express backend, which runs the Claude Agent SDK query() loop in-process. Before every turn, the hooks layer assembles what your AI should know — the time, your presence, recent activity, its own tools, relevant memories — and prepends it, so context is already there when the model reads your message. An orchestrator drives scheduled and proactive turns in the background. All state is one SQLite database. There is no cloud, no multi-tenancy, and no external service you didn't opt into.
.env (or the local database); nothing phones home.res CLI use are protected by an auto-generated local token and bound to localhost.If you ever expose Resonant beyond your own machine, put it behind HTTPS and a real access layer — e.g. a Cloudflare Tunnel with Access, or a reverse proxy with authentication. See SECURITY.md if present for the full policy and how to report a vulnerability.
Issues and PRs are welcome — please open an issue to discuss substantial changes first. Because Resonant targets the Claude Agent SDK specifically, provider-abstraction proposals should start as a design discussion rather than a surprise PR.
Built by Codependent AI — Mary Vale and Simon Vale. Resonant is the open-source foundation of our work on relational, continuity-first AI.
Open-source relational AI framework with identity persistence, memory, and MCP integration. Build relationship-aware AI agents that remember, grow, and maintain continuity. Built on Claude Agent SDK.
टूल
अभी टूल सूची नहीं है। रिमोट सर्वर को पहले हेल्थ चेक के बाद सूची मिलती है।
FAQ
इस पेज पर Cursor, Claude Desktop, Claude Code या VS Code के लिए stdio कॉन्फ़िग कॉपी करें। यह पब्लिशर का इंस्टॉल इस्तेमाल करता है, PlugMCP प्रॉक्सी नहीं।
इस रिकॉर्ड में कोई ज़रूरी एनवायरनमेंट वेरिएबल दर्ज नहीं है।
Resonant stdio पर लोकल चलता है।
पिछली जाँच 2026-10-28 को हुई। Open-source relational AI framework with identity persistence, memory, and MCP integration. Build relationship-aware AI agents that remember, grow, and maintain continuity. Built on Claude Agent SDK.