Trading agents,with guardrails.
A TypeScript harness for traders who want automation without handing execution to an unchecked bot. It reads portfolios, watches markets, simulates routes, proposes trades, and keeps live execution behind explicit approval gates.
$ npm install -g zap-agent@latest$ zap init$ zap agents$ zap run --agent news_agent --dry-run
Built in the same local-harness spirit as OpenClaw and Hermes: inspectable tools, configurable providers, and trader-owned control before execution.
[ Core Harness ]
Trading features wrapped in a local control loop.
z4p is not a chatbot wrapper. It is the runtime around the agent: data intake, memory, policy, proposals, approvals, execution, and audit logs.
The default posture is observe-only and dry-run. Real trading requires configured wallets, limits, and a trader approval path.
Ingest
Read wallets, balances, open positions, market data, news, research feeds, venue state, and configured risk limits.
Analyze
Run specialist agents for research, portfolio review, route checks, strategy specs, and trade rationale.
Simulate
Compare Jupiter, Uniswap, Hyperliquid, and configured venues for slippage, fees, liquidity, and post-trade exposure.
Approve
Keep live execution behind trader review, risk limits, allowance-backed approvals, halt controls, and audit logs.
Execute
Submit approved orders, track fills, update positions, record outcomes, and feed the result back into memory.
Solana trading with SPL balances, Jupiter quotes, proposal tickets, and supervised order flow.
EVM and Uniswap flows for account reads, ERC-20 allowances, quote checks, approvals, and swaps.
Hyperliquid access for info, exchange actions, subscription streams, positions, and managed order workflows.
Portfolio manager, news, research, strategy builder, Solana/Jupiter, Pump.fun, and Uniswap agent templates.
Paper-first defaults, dry-run commands, global halt controls, managed orders, outcome tracking, and audit logs.
OpenAI, Anthropic, Gemini, Copilot, OpenClaw/Hermes-style OAuth flows, and custom provider gateways.
Harness architecture
local firstAgent runtime
Schedules agent ticks, builds prompts, coordinates specialist agents, and routes tool calls through one harness.
Memory and journals
Stores sessions, quotes, positions, proposals, decisions, facts, feedback, fills, and maintenance tasks locally.
Policy layer
Applies risk limits, allowances, approval requirements, stale quote checks, pause states, and global halts.
Execution adapters
Connects to Solana, Jupiter, EVM, Uniswap, Hyperliquid, research sources, webhooks, Telegram, MCP, and metrics.
[ Quickstart ]
Add the harness and run an observe-only agent.
Start with dry-run workflows. The harness can read context, build proposals, and show what it would do before you configure live wallets or approval-backed execution.
Install globally
Fastest path for traders.
$ npm install -g zap-agent@latest$ zap init$ zap agents$ zap run --agent news_agent --dry-run
Verify install
Confirm the command is on your path.
$ zap doctor$ zap agents
Run from source
Best path for contributors.
$ git clone https://github.com/zapmarkets/zap_.git$ cd zap_$ pnpm install$ pnpm run build$ pnpm exec tsx src/cli.ts doctor
Useful first commands
observe firstzap run --agent portfolio_manager --dry-runzap run --agent research_agent --dry-runzap trade:list --status proposedzap trade:approve --id <proposal-id>[ What Ships ]
Built for traders, not demos.
The harness gives agents practical surfaces to work through while keeping the trader close to risk, approvals, and execution.
CLI
Initialize config, inspect wallets, run agents, review proposals, approve trades, and generate reports from one command surface.
Local UI
Run the browser interface for sessions, trading agents, control-plane work, status, and trader review.
MCP tools
Expose readiness, context, agent ticks, halt status, and trading workflows to compatible coding and agent environments.
Automations
Use Telegram, webhooks, metrics, backups, scheduler jobs, and daemon workflows when the harness needs to stay online.
[ Companion App ]
The app is where the trader stays close.
z4p is the mobile companion for the harness: portfolio, markets, alerts, approvals, and control when the agent has something worth your attention.
Approval surface
Review proposed actions, risk notes, fills, and emergency controls from the app instead of giving the agent unattended custody.
Portfolio context
Use the app for balances, markets, swaps, perps, alerts, and the day-to-day view of what the harness is watching.
Human in the loop
The harness can prepare and explain actions; sensitive execution still routes through local signing and explicit trader control.