Deploy an agent through the web UI, configure its instructions and control profile, and start a conversation.

Your First Agent

This guide walks you through deploying your first agent in the web app, from strategy selection through your first Chat conversation.

Prerequisites

Before you start, you need:

A Cabal account at
At least one model provider connected (see )
Onboarding completed -- you should be able to reach /dashboard

Open the deploy flow

Go to Agents in the dashboard sidebar, or navigate directly to /dashboard/agents. If you have no agents yet, the page shows an empty state with a Deploy agent button. Click it to open the deploy wizard at /dashboard/hire.

The wizard has five steps: Strategy, Identity, Instructions, Readiness, and Deploy.

Step 1: Choose a strategy

Templates seed your agent's Standing Instructions with a starting point you can edit. Pick the one closest to how you want the agent to operate.

Template	What it does
Momentum scout	Tracks breakouts, relative strength, and follow-through. Sizes conservatively and explains invalidation before proposing.
LST watcher	Monitors staking, liquid-staking, and yield dislocations. Prefers evidence-backed observations.
Research desk	Builds theses from market, token, and protocol evidence. Separates thesis, evidence, risks, and trade setup.
Custom agent	Starts blank. Write your own instructions from scratch.

You can change the instructions in the next step regardless of which template you pick.

Step 2: Set the identity

Give your agent a display name and a handle. The handle is the public identifier (like @sol-desk) and must be at least 3 characters -- lowercase letters, numbers, underscores, or hyphens. Cabal checks availability in real time and reserves the handle when you continue.

Step 3: Seed Standing Instructions

The instructions editor is pre-filled from your chosen template. Edit them to match your preferences. Instructions describe judgment and preferences -- what to watch, what to avoid, what to explain before acting. They are guidance, not enforcement. Hard stops belong in Guardrails.

Instructions must be at least 20 characters before you can continue.

Tips for writing good instructions:

Be specific: "SOL majors only" beats "be careful."
State preferences, not rules. Rules belong in Guardrails.
Include timeframes: "no opens after 22:00 ET."
Avoid vague guidance: "be smart about risk" tells Cabal nothing.

Step 4: Check readiness

The readiness step checks three things before deployment:

Model provider -- Whether you have a provider connected and which model it defaults to.
Wallet custody -- Whether you have a wallet linked. Required for Guarded and Full Access. Optional for Suggest Only.
Guardrail defaults -- Whether Cabal's default guardrail settings are ready to save on the agent.

This step also asks you to pick a control posture for the agent:

Posture	What it means
Suggest Only	Cabal proposes trades. You approve each one. Best for new users or large portfolios.
Guarded	Cabal executes on its own, but only within your saved guardrails -- position sizes, venue restrictions, and other limits.
Full Access	Cabal executes broadly. Guardrails stay on file but are not actively enforced.

Start with Suggest Only unless you have a specific reason not to. You can switch at any time from the agent's Control tab.

Step 5: Deploy

The final step shows a summary: identity, template, posture, and the status of each readiness check. If all checks pass, click Deploy agent. Cabal creates the agent and redirects you to its page at /dashboard/agents/<handle>.

After deployment: the agent page

Your agent page has four tabs:

Overview

Shows aggregate metrics (market value, unrealized PnL, realized PnL, open positions), a positions table, and a recent activity feed. On a new agent, these are empty until the agent starts trading.

Control

Where you switch between Suggest Only, Guarded, and Full Access. When Guarded is active, the guardrails editor appears below the mode cards. You can set position size limits, venue restrictions, slippage tolerance, and other execution constraints. Recent guardrail refusals are listed at the bottom.

Instructions

The freeform text editor for Standing Instructions. Cabal reads these on every prompt. The editor tracks character and line counts, supports save and discard, and keeps a version history with rollback.

Bindings

Controls which account-level resources this specific agent can use:

Model -- Pick the provider and model this agent defaults to. Different agents can use different models.
Wallets -- Bind a Solana or Ethereum wallet to the agent for trade execution. Wallets are linked at the account level in Settings; bindings control which ones this agent uses.
Extensions -- Enable or disable Cabal-curated capabilities for this agent.

Start a conversation

Open Chat from the dashboard sidebar. Select your new agent from the agent selector if you have more than one. Type a message -- ask a question, request research, or describe a trade idea.

If your agent is set to Suggest Only, any trade proposal appears as a structured action card in the conversation. You review the reasoning, slippage estimate, and venue, then approve or reject.

If your agent is set to Guarded, it can execute within your guardrails without asking. Trades that exceed limits are blocked and logged in the Activity feed and the Control tab's refusals list.

What to try first

Ask your agent to research a token: "What is the current state of SOL?"
Ask for a trade idea: "Any momentum setups on Solana right now?"
Review the reasoning before approving: the agent explains its thesis, evidence, and risk assessment inline.
Check the Activity feed after a trade to see the full execution record.