# Cortex Agent Builder > The Agent Builder is where you create a custom Cortex agent. Open it from the Cortex section in the web app (/cortex/builder). The flow is four steps - same… Source: https://tradeecho.com/docs/features/cortex/agent-builder (updated 2026-09-26) The **Agent Builder** is where you create a custom Cortex agent. Open it from the Cortex section in the web app (`/cortex/builder`). The flow is four steps - same labels you see in the app: 1. **Pick a base** 2. **Teach it** 3. **Set the rails** 4. **Publish** You can click back to any completed step without losing your work. Custom agents require an active Trade Echo subscription. If you only need chat and presets, see [Cortex Quick Start](https://tradeecho.com/docs/features/cortex/quick-start). --- ## Step 1: Pick a base Choose one of **18 built-in bases** (presets). The base sets which market-data tools the agent prefers - for example flow-heavy vs GEX-heavy. - Browse the preset cards in Step 1 and pick the closest match to your strategy. - Not sure? Start with **Full Scan** for maximum tool access, or match the preset you already use in Cortex chat. The full preset table lives on [Cortex Quick Start → Built-In Presets](https://tradeecho.com/docs/features/cortex/quick-start#built-in-presets). --- ## Step 2: Teach it This is your strategy layer on top of the base. ### Custom instructions - Up to **2,000 characters** of rules: setups to take, setups to skip, selectivity, exit logic, and tone. - Copy-paste starters: [Ready-Made Custom Prompts](https://tradeecho.com/docs/features/cortex/ready-made-prompts). ### Rule builder Define scan conditions (for example ticker, session window, or signal thresholds) and **test the rule against the latest market snapshot** before you save. Testing confirms the rule matches real data - not just text on a page. ### Chat composer (optional) Step 2 also offers a **chat composer** mode: describe your strategy in plain language and let Cortex propose instruction edits. Switch back to the form view anytime to fine-tune fields by hand. --- ## Step 3: Set the rails The trust step. Everything here defines what the agent is allowed to do. ### Execution mode | Mode | What it does | |------|----------------| | **Research only** | Reads the market and surfaces signals. No trades - paper or live. | | **Paper trade** | Simulates trades in a paper book using your rules. **Recommended to start.** | | **Live execution** | Routes real orders through your linked broker. Requires confirmation before save. | **Live** in the Builder (your **own** agent on your broker): | Membership | Own-agent live | |------------|----------------| | **Monthly / yearly** | **1 live agent** at a time. That one live slot is shared: it can be your own agent **or** one live [Marketplace follow](https://tradeecho.com/docs/features/cortex/marketplace#subscribe-to-a-listing), not both. Your live agent trades on your **active Copy Trading account**; you cannot pick a different broker per agent. | | **Lifetime** | **Unlimited** live agents, your own and followed. You can **pick a broker per agent**. | Paper agents and paper follows never use the live slot. To make a different agent live on a monthly or yearly plan, first move your current live agent back to Paper (Sim) or unfollow the live Marketplace agent. Before 2026-09-26, running your own agent live needed Lifetime; monthly and yearly members can now do it too. ### Approval mode - **Full auto** - trades within your rails without asking each time. - **Review mode** - every idea queues in [Approvals](https://tradeecho.com/docs/features/cortex/agents#approvals) until you approve or it expires. Start with **review mode** until you trust the agent's behavior. ### Broker binding (live only) Live agents need a **validated broker connection** bound in Trade Echo. If the broker session drops, the agent pauses with **Broker offline** until you reconnect. - **Monthly / yearly:** the broker card is locked to your **active Copy Trading account** (set it with **Set Active** on the Copy Trading **Broker Accounts** tab). If you have no active account, the card points you to Copy Trading first. - **Lifetime:** pick any linked account for each agent. - You cannot switch the broker on an agent while it has open live trades. ### Risk rails Typical controls on this step: - **Sizing model** - how the agent sizes against paper balance or live account context - **Starting balance** (paper) - simulation baseline - **Allowed instruments** - equity, options, or both - **Allowed expirations** - 0DTE, weekly, monthly - **Risk envelope** - max positions, max daily orders, max dollars per trade, max contracts per intent, execution timeout - **Drawdown kill-switch** - automatic pause on daily loss limits - **Weekly schedule** - which days and session windows the agent runs, in your timezone - **Trading calendar** - two optional switches that narrow the schedule: - **Trade only during US market hours** keeps the agent to NYSE trading days, 9:30 ET to the close (1:00 PM ET on early-close days such as the day after Thanksgiving and Christmas Eve). - **Skip US market holidays** keeps the agent idle on NYSE holidays such as Good Friday. Columbus Day and Veterans Day are normal trading days. ### Exits The **Exits** card controls how every trade closes. Settings here are enforced by the server after the fill - instructions in Step 2 guide the AI's plan, but they never sell anything on their own. | Setting | What it does | |---------|--------------| | **Profit target** | The gain (% of the entry premium) the agent aims for on each trade | | **Hard stop** | The largest loss (% of the actual entry fill) the position may take before it closes. Leave blank to use the default option minimum | | **Trail after target** | Instead of closing at the target, the server trails the position and locks in a floor (**Never give back below**) so winners can run | | **Take partial profit** | Up to **3 steps** of "sell X% at +Y%" - each step sells that share of the position the agent opened when the gain is reached. The rest keeps its stop, target and trail | How partial profits work: - Each step sells a share of the **original** position. A step never sells the last contract unless your steps add up to **100%**. - A **1-contract** position sells in full at the **first** step. - **Marketplace subscribers** follow the same steps on their own position size. If the agent's position closes, subscribers close too. - **Live** agents sell their 1 contract at the first step. - Writing "take half off at +30%" in your instructions does **not** trim. Set it here. The **Decisions** tab flags agents whose instructions mention trimming without steps saved. - Backtests model your steps, and **Why it exited** lists each trim on a closed trade. ### Authority summary Before you save, the **authority summary** lists exactly what you authorized - mode, caps, schedule, and approval behavior. Read it before continuing to Publish. ### Paper history (live) Before you switch your **own** agent to **live**, the Builder shows how many days and trades it has on paper. Most members run an agent on paper for **at least 7 days** first. This is a recommendation, not a lock: on any paid plan you can enable live once the broker, rails, and **ENABLE LIVE** confirmation are complete, and the Builder shows a banner while the agent is short of 7 paper days. Paper mode is always available while you build history. --- ## Step 4: Publish Name the agent, choose visibility, and optionally list it on the Marketplace. ### Agent name A clear name helps you find the agent under **My Agents** later. ### Public description (community listings) Up to **500 characters** for marketplace listings - what the strategy does, instruments, and session style. ### Tags Pick tags for **strategy**, **instruments**, and **vibe** so members can filter listings. Tags auto-suggest from earlier steps; you can refine them here. ### Visibility | Option | Meaning | |--------|---------| | **Private** | Only you see and run this agent. | | **Community** | Published to the [Marketplace](https://tradeecho.com/docs/features/cortex/marketplace). Others can subscribe and follow signals. | Community listings show a **data transparency** panel before you publish - what performance and config details followers can see. ### After you save - **Private** agents appear under [My Agents](https://tradeecho.com/docs/features/cortex/agents). - **Community** agents redirect to the Marketplace **My listings** tab, where followers can subscribe. To edit later, open the agent from **My Agents** or **My listings** and return to the Builder. --- ## Paper vs live (creators) - Every new agent defaults to **paper** unless you choose otherwise. - The platform **never** promotes an agent to live automatically. - **Monthly / yearly:** enable live in the Builder when ready. One live agent at a time (shared with a live Marketplace follow), on your active Copy Trading account. - **Lifetime:** enable live in the Builder when ready, with no live agent limit and a broker picked per agent. - Paper fills, slippage, and timing do not match real broker execution. Judge paper results as a simulation, not a guarantee of live P&L. Cortex agents are research and automation tools, not personalized investment advice. --- ## Next steps - [Marketplace](https://tradeecho.com/docs/features/cortex/marketplace) - browse, list, and subscribe to community agents - [My Agents and Approvals](https://tradeecho.com/docs/features/cortex/agents) - monitor runs, paper book, and review queue - [Ready-Made Custom Prompts](https://tradeecho.com/docs/features/cortex/ready-made-prompts) - paste-ready Step 2 instructions