/docs
Install
Guided Installer

Install Spark from the visual tutorial.

The main install experience now lives at /install. Use that page first, then come back here when you want the same path with more detail.

Start Install Model Setup Daily Commands
Step 1

Tell Codex, Claude Code, or your setup assistant to install Spark.

The easiest path is to ask an AI coding assistant like Codex or Claude Code to read agent.sparkswarm.ai/install and install Spark with you. They can explain what is happening as they go.

  • The assistant picks the right Mac, Linux, WSL, or Windows install path.
  • It can run the dry run first, so you can check whether the download and installer look secure before installing.
  • It should help you choose one model provider before asking you to connect Telegram.

You can still install from terminal by hand, but non-technical users usually have a much smoother time with a coding assistant guiding the install.

Step 2

Choose the LLM Spark will use.

Start simple: choose one default LLM provider for Spark. That can power Agent chat, memory, runtime reasoning, and Mission work while you are getting started.

  • Agent means normal chat, reasoning, preferences, memory, and recall.
  • Mission means Mission Control builds, research, coding, and longer tracked work.
  • You can use one provider for both, or split Agent and Mission during setup.
  • Spark chooses current recommended defaults for a provider; use the provider docs if you want exact model names.
  • If you already have a Codex, Claude Code, Cursor, or another local agent subscription, tell it to connect that path with Spark.
  • Use an API key if you prefer Z.AI GLM, Kimi, MiniMax, Gemma, OpenRouter, or DeepSeek-compatible routes.
  • Use local models through Ollama, LM Studio, Hugging Face, or Gemma when your computer is strong enough.
Step 3

Create your Spark Telegram bot.

Open Telegram, search for @BotFather, send /newbot, and follow the prompts. BotFather gives you a bot token. If you already have a bot, you can use that token instead.

  • Spark uses that bot as the chat door to your agent.
  • Your own Telegram account becomes the admin account, so only you can run powerful Spark commands.
  • Paste the token only into local Spark setup. Never paste it into the website.
Abstract green and purple Spark learning loop
Build resonance Spark learns your taste, habits, and working style so each conversation starts closer to you.
Step 4

Start chatting and build resonance.

Before building, just talk to Spark. Tell it how you like to work, what you care about, what you dislike, and what you want help remembering.

  • Use normal chat for daily ideas, plans, and questions.
  • Use /remember for preferences Spark should keep.
  • Use /recall to check what Spark remembers.
  • Use /diagnose if the bot feels quiet or confused.
Step 5

Build with Spark Mission Control.

When you are ready, ask Spark to build something small. Mission Control shows the work through Kanban and Canvas so you can see what is happening.

  • Start with a tiny app, workflow, or tool.
  • Then build domain chips when Spark needs to master a specific topic or skill.
  • Use autoloops when you want Spark to keep improving a repeated process over time.
  • The Domain Chip Creator path helps turn useful know-how into reusable Spark skills.

Always-on hosting comes after the local path.

Finish the local install before touching hosting. Railway, Docker, Spawner, private service links, and persistent volumes are advanced operations details, so they now live on the dedicated operations page instead of the beginner installer path.

Local first

Install, choose a provider, connect Telegram, and verify onboarding before deploying anything.

Hosted later

Use the Railway/VPS page when you need persistent hosting, service wiring, and smoke tests.

Keep secrets local

Beginner setup should never ask you to paste tokens or bridge keys into a website.

Each user should normally run hosted Spark in their own hosting account so costs, workspaces, and secrets stay under their control.

Use the Railway/VPS operations page when you need the full hosted service map, environment checklist, and production smoke prompt.

A Spark that keeps getting better.

Spark is meant to become a personal system that learns your taste, your projects, and the tools you care about. With the right domain chips and benchmarks, it can keep practicing a skill or workflow even while you sleep. You get to see that progress through memory, visible missions, and recursive self-improvement loops.

Mission board

Users can watch work through Kanban and Canvas instead of guessing what an agent is doing.

Memory

Spark can save useful preferences and facts, then use them in later conversations and missions.

Recursive self-improvement

Spark can create domain chips and tools to master repeatable work, then reuse them on the next loop.