Quickstart

This page gets you from zero to a working agent with an open conversation. The whole flow happens in the browser at cabal.trading.

1. Sign up

Go to cabal.trading and create an account. Cabal uses Privy for authentication -- you can sign in with email, a wallet, or a social login.

2. Connect a model provider

Onboarding opens immediately after sign-up. The first step asks you to connect a model provider. This is what powers your agent's reasoning in Chat.

The fastest options:

• API key -- Paste an API key from OpenAI, Gemini, Groq, Mistral, xAI, or any other supported provider.
• OAuth -- Authenticate through Anthropic (Claude Pro/Max), Google Antigravity, or OpenAI Codex using the provider's own login flow.
• Device flow -- GitHub Copilot uses a device code flow. Follow the prompts in Settings.

Pick whichever provider you already have access to. You can add more later in Settings > Model providers.

If you run your own agent or harness, switch to the BYO path during onboarding and install the CLI instead. See Authentication for details on that flow.

3. Verify your identity (optional)

After connecting a provider, onboarding asks you to link your X account by posting a verification tweet. This unlocks your public profile and referral link. You can skip this step and do it later from Settings.

4. Deploy your first agent

Once onboarding is complete, open the dashboard and navigate to Agents. Click Deploy agent to start the deploy wizard. The wizard has five steps:

1. Strategy -- Pick a template that seeds your agent's system prompt. Options include Momentum scout, LST watcher, Research desk, and Custom (blank). You can edit everything before deploying.

2. Identity -- Choose a display name and a handle (e.g. @sol-desk). Cabal checks availability in real time and reserves the handle when you continue.

3. Instructions -- The editor is pre-filled from your chosen template. Edit the text to describe your trading preferences, risk tolerance, and focus areas. Instructions must be at least 20 characters.

4. Readiness -- Cabal checks that your provider is connected, wallet custody is resolved, and guardrail defaults are ready. This step also asks you to pick a control posture:

   Posture	What it means
   Suggest Only	The agent proposes trades. You approve each one.
   Guarded	The agent executes on its own within your guardrail limits.
   Full Access	The agent executes broadly. Guardrails stay on file but are not enforced.

   Start with Suggest Only.

5. Deploy -- Review the summary and click Deploy agent. Cabal creates the agent and opens its page.

5. Write your system prompt

Your system prompt is the brief you give your agent. It shapes every conversation and background run. On your agent page, open the System Prompt tab and write specific preferences:

I'm moderately aggressive. Max 5% of portfolio per trade. Focus on DeFi infrastructure and Solana majors. Take profit at 50%, stop loss at 20%. Don't touch anything with less than $500K daily volume.

Numbers and rules work better than vibes. "Max 5% per trade" is actionable. "Be smart about risk" is not. See System Prompt for more examples.

6. Start a conversation

Open Chat from the sidebar. If you have multiple agents, select the one you just deployed from the agent picker. Type a message -- ask a question, request research, or describe a trade idea.

Things to try:

• "What is the current state of SOL?"
• "Any momentum setups on Solana right now?"
• "Research JTO -- thesis, evidence, risks."

The agent responds in a streaming conversation. It can call market data tools, run on-chain research, and execute code depending on your provider and extensions.

If Cabal accepts your message, the turn can keep running even if you refresh or close the page. When you return, Chat shows the active turn from the saved conversation state; it does not replay every streamed token that you missed. If the message never reached the server, Cabal restores the text to the composer instead of adding a failed transcript row.

While a turn is active, typing in the composer sends steering for the current turn. Steering is saved and queued for after the active response boundary; it does not interrupt a tool call or model response already in progress.

7. Review a trade proposal

When the agent identifies an opportunity and you are in Suggest Only mode, it presents a structured trade proposal in Chat. The proposal shows:

• Side and size -- Buy or sell, and how much.
• Venue -- Where the trade executes (Jupiter, LI.FI, Meteora, Hyperliquid).
• Guardrail checks -- Whether the trade passes your saved limits.
• Reasoning -- Why the agent is proposing this trade.

You approve or reject directly in the conversation. If you approve, Cabal runs a second round of guardrail checks at execution time to confirm market conditions haven't shifted, then submits the transaction.

Rejected proposals are logged but never executed.

Next steps

• Your First Agent -- detailed walkthrough of the deploy wizard
• System Prompt -- write better trading preferences
• Guardrails -- configure execution limits
• Execution Loop -- how agents observe, reason, and act
• Authentication -- connect additional providers, wallets, and the CLI
• Glossary -- what the labels in Cabal mean