From signup to first agent turn in five minutes.

1. Create a workspace

Go to Register, pick an email and a password. Your account is the tenant; every query on the server filters by your tenant id. You cannot see anyone else's data and nobody can see yours.

2. Connect a provider

Go to Settings → Providers and pick one of the four supported backends.

• OpenAI — paste a sk-… key. Pick a model like gpt-4o-mini or gpt-4o.
• Anthropic — paste a sk-ant-… key. Pick a Claude model (e.g. claude-sonnet-4-5, claude-haiku-4-5).
• OpenRouter — paste an sk-or-… key. Pick any model id from the OpenRouter catalog.
• Ollama — run ollama serve locally, then expose it with a tunnel (Cloudflare Tunnel or ngrok) so the hosted GnamiAI server can reach it. Paste the tunnel URL as the base URL and pick a model you have pulled (e.g. llama3.1). Inference still runs on your machine — the tunnel only routes the request. Loopback and private IPs are rejected by the SSRF guard.

Keys are encrypted with AES-256-GCM before they hit the database. The server holds the key-encryption key; a database dump alone does not leak your credentials.

3. Run your first chat

Go to Chat and type a prompt. The agent calls your selected provider, records the turn in your budget ledger, and shows you context it used (memory hits, skills loaded, active subagent).

Useful slash commands you can type in the composer:

• /help — list every command.
• /clear — wipe the local transcript.
• /models — show the catalog for your active provider.
• /model <id> — switch model for your provider.
• /budget — see today's token and USD spend.
• /remember <fact> — write a long-term memory to Mem0.
• /install <url> — install a SKILL.md from a URL.
• /agent <name> — pin a subagent for the rest of the session.

4. Create a subagent

A subagent is a named specialization with its own system prompt and (optionally) a preferred model. Go to Subagents, hit New, and describe the agent's role. Pin it in chat with /agent <name> and every turn will use that agent's prompt stack.

5. Install a skill

A skill is a plain SKILL.md that tells the agent how to do something (e.g. "write conventional-commit messages", "draft release notes"). On Skills you can create one by hand or install from a URL. URL installs are fetched server-side with SSRF guards, classified to reject HTML/binary, and NUL-byte-scrubbed before storage.

6. Connect long-term memory (optional)

In Settings → Memory paste a Mem0 API key and, optionally, an entity_id, org_id, and project_id. Default entity is gnamiai:tenant:<your-id>. The agent lists your memories at turn start and can write new ones with /remember or a structured action.

7. Schedule a recurring run

Go to Schedules, create a new cron with a name, an interval (minimum 24h on Hobby), and the prompt you'd type. The tick runs once a day and executes every due schedule; each run stores its output, any actions taken, and status.

Ready?

Create a workspace and run a turn. It takes longer to read this page than to try it.