Skip to content

api-reference.md

Latest commit

 

History

History
1052 lines (772 loc) · 26.2 KB

api-reference.md

File metadata and controls

1052 lines (772 loc) · 26.2 KB
Raw

API Reference — Pulling Data from Nunchi Agents

This guide covers every method for pulling data from a Nunchi-hosted agent. Hosted agents are provisioned only through web-auth after billing entitlement and wallet binding.

Path Protocol Best For
HTTP REST API HTTP/JSON Dashboards, monitoring, external integrations
SSE Feed Server-Sent Events Live streaming to frontends
MCP Server Model Context Protocol AI agent orchestration (Claude, OpenClaw)

Security: Hosted deployment is gated by web-auth billing entitlement and wallet binding. Mutating control endpoints on the hosted runner require operator-provisioned control tokens; users should interact through web-auth and the scoped hosted MCP flow instead of direct infrastructure access.

Prerequisites

Agent Deployment

Deploy hosted agents from web-auth:

https://auth.nunchi.trade

The web-auth flow handles billing, wallet binding, provisioning through Nunchi's Railway account, secret injection, and hosted endpoint discovery. This repository does not publish public Docker/Railway deployment templates.

Base URL

Throughout this document, $AGENT_URL refers to your agent's base URL:

# Hosted endpoint returned by web-auth after deployment
export AGENT_URL=https://your-agent-hosted-endpoint.example

# Verify connectivity
curl $AGENT_URL/health

HTTP REST API

GET /health

Health check endpoint. Use this to verify the agent is reachable and the trading process is alive.

curl $AGENT_URL/health

Response:

{
  "status": "ok",
  "mode": "apex",
  "uptime_s": 3842,
  "pid": 127,
  "alive": true
}
Field Type Description
status string Always "ok" if server is responding
mode string RUN_MODE env var: "apex", "strategy", or "mcp"
uptime_s int Seconds since the entrypoint started
pid int|null PID of the child trading process
alive bool Whether the child process is still running

Notes:

  • Used by Nunchi's hosted-agent provisioner and health monitoring.
  • Returns 200 if the HTTP server is responding; check alive to distinguish a live trading process from an unhealthy child process.

GET /api/status

Primary status endpoint. Returns the agent's current state, positions, and PnL.

curl $AGENT_URL/api/status

Response (APEX mode):

{
  "status": "running",
  "engine": "apex",
  "tick_count": 284,
  "daily_pnl": -12.50,
  "total_pnl": 342.00,
  "total_trades": 28,
  "max_slots": 3,
  "active_slots": [
    {
      "slot_id": 0,
      "instrument": "VXX-USDYP",
      "side": "long",
      "entry_size": 2.5,
      "entry_price": 30.552,
      "roe_pct": 4.2,
      "dsl_phase": 2,
      "status": "active"
    }
  ],
  "closed_slots": [],
  "positions": [
    {
      "slot": 0,
      "market": "VXX-USDYP",
      "side": "long",
      "size": 2.5,
      "entry": 30.552,
      "roe": 4.2,
      "phase": 2
    }
  ]
}

Response (single-strategy mode):

{
  "status": "running",
  "engine": "engine_mm",
  "tick_count": 1042,
  "instrument": "ETH-PERP",
  "position_qty": 0.5,
  "unrealized_pnl": 3.42,
  "realized_pnl": 18.90,
  "total_orders": 312,
  "total_fills": 198
}

Response (agent not running):

{
  "status": "stopped"
}
Field Mode Type Description
status Both string "running" or "stopped"
engine Both string Strategy ID ("apex", "engine_mm", etc.)
tick_count Both int Number of evaluation cycles completed
daily_pnl APEX float Today's profit/loss in USD
total_pnl APEX float Cumulative profit/loss in USD
total_trades APEX int Number of executed trades
max_slots APEX int Maximum concurrent positions
active_slots APEX array Currently open position details (full slot objects)
closed_slots APEX array Last 5 closed positions
positions APEX array Simplified position array for UI consumption
instrument Strategy string Trading instrument (e.g., "ETH-PERP")
position_qty Strategy float Net position quantity (positive = long)
unrealized_pnl Strategy float Mark-to-market PnL
realized_pnl Strategy float Locked PnL from closed trades
total_orders Strategy int Orders placed this session
total_fills Strategy int Orders that filled

Data source: Reads from $DATA_DIR/apex/state.json (APEX mode) or $DATA_DIR/cli/state.db (strategy mode). These files are written by the trading engine after every tick.

Fallback behavior: If the Python status reader fails, the Node.js server reads state.json directly as a fallback. The Python entrypoint has no fallback.

GET /api/strategies

Returns the full catalog of available trading strategies and YEX markets.

curl $AGENT_URL/api/strategies

Response:

{
  "strategies": {
    "simple_mm": {
      "description": "Fixed-spread market maker",
      "params": { "spread_bps": 10, "base_size": 0.5 }
    },
    "avellaneda_mm": {
      "description": "Avellaneda-Stoikov inventory-aware market maker",
      "params": { "gamma": 0.1, "k": 1.5, "base_size": 0.5, "max_inventory": 5 }
    },
    "engine_mm": { "..." : "..." },
    "regime_mm": { "..." : "..." },
    "grid_mm": { "..." : "..." },
    "liquidation_mm": { "..." : "..." },
    "funding_arb": { "..." : "..." },
    "basis_arb": { "..." : "..." },
    "momentum_breakout": { "..." : "..." },
    "mean_reversion": { "..." : "..." },
    "aggressive_taker": { "..." : "..." },
    "hedge_agent": { "..." : "..." },
    "rfq_agent": { "..." : "..." },
    "claude_agent": { "..." : "..." }
  },
  "markets": {
    "VXX-USDYP": "Volatility index yield perpetual",
    "US3M-USDYP": "US 3-month Treasury rate yield perpetual",
    "BTCSWP-USDYP": "BTC interest rate swap yield perpetual"
  }
}

Each strategy includes its description, type, and params with default values. Use this to populate strategy selection UI or validate configuration inputs.

GET /status

Human-readable plain-text status. Calls hl apex status internally.

curl $AGENT_URL/status

Response:

APEX Orchestrator — default preset
Tick: 284 | Daily PnL: -$12.50 | Slots: 1/3

Slot 0: VXX-USDYP LONG 2.5 @ 30.552 | ROE: +4.2% | Phase: Tier 2
Slot 1: (empty)
Slot 2: (empty)

Same output as running hl apex status in a terminal. Useful for quick checks but not suitable for programmatic consumption — use /api/status instead.

POST /api/skill/install

Verifies that the agent has the Nunchi trading CLI installed and returns the strategy count.

curl -X POST $AGENT_URL/api/skill/install \
  -H "Content-Type: application/json"

Response:

{
  "installed": true,
  "strategies": 14,
  "tools": 13
}

Error response:

{
  "installed": false,
  "error": "ModuleNotFoundError: No module named 'cli'"
}

Use this as a connectivity + capability check before wiring a UI to the agent.

POST /api/pause

Pauses the trading process by sending SIGSTOP to the child process. The agent stops executing ticks but maintains all state. Positions remain open.

curl -X POST $AGENT_URL/api/pause \
  -H "Authorization: Bearer $API_AUTH_TOKEN"

Response:

{ "status": "paused" }

Error (no process running):

{ "error": "No running agent to pause" }

Error (token not configured):

{ "error": "control_auth_required" }

Warning: Pausing stops the DSL trailing stop from updating. If the market moves significantly while paused, positions will not be protected.

POST /api/resume

Resumes a paused trading process by sending SIGCONT.

curl -X POST $AGENT_URL/api/resume \
  -H "Authorization: Bearer $API_AUTH_TOKEN"

Response:

{ "status": "resumed" }

CORS

All /api/* endpoints return CORS headers:

Access-Control-Allow-Origin: * (or $CORS_ORIGIN env var)
Access-Control-Allow-Methods: GET, POST, OPTIONS
Access-Control-Allow-Headers: Content-Type, Authorization, X-API-Token

OPTIONS requests to any path return 204 with these headers. Set the CORS_ORIGIN environment variable to restrict origins in production.

SSE Real-Time Feed

GET /api/feed

Server-Sent Events stream that pushes the agent's status every time the tick counter changes. The server polls the state file every 2 seconds and emits an event only when tick_count has advanced.

JavaScript:

const source = new EventSource(`${AGENT_URL}/api/feed`);

source.onmessage = (event) => {
  const data = JSON.parse(event.data);
  console.log(`Tick ${data.tick_count}:`, data);
};

source.onerror = () => {
  console.log('SSE connection lost, will auto-reconnect');
};

curl (for testing):

curl -N $AGENT_URL/api/feed

Event format:

data: {"status":"running","engine":"apex","tick_count":285,"daily_pnl":-11.20,...}

data: {"status":"running","engine":"apex","tick_count":286,"daily_pnl":-10.80,...}

Each data: line contains the same JSON payload as GET /api/status. Events are separated by double newlines per the SSE specification.

Headers returned:

Content-Type: text/event-stream
Cache-Control: no-cache
X-Accel-Buffering: no

The X-Accel-Buffering: no header prevents Nginx reverse proxies from buffering the stream.

Behavior:

  • Emits immediately on first connect (current state)
  • Then emits only when tick_count changes (de-duplicated)
  • APEX ticks every 60 seconds by default, so events arrive roughly every 60s
  • Single-strategy ticks are configurable (default 10s), so events arrive more frequently
  • The connection stays open indefinitely until the client disconnects
  • EventSource auto-reconnects on connection loss (browser default behavior)

Python example:

import json
import requests

response = requests.get(f"{AGENT_URL}/api/feed", stream=True)
for line in response.iter_lines():
    if line:
        text = line.decode("utf-8")
        if text.startswith("data: "):
            payload = json.loads(text[6:])
            print(f"Tick {payload['tick_count']}: PnL ${payload.get('daily_pnl', 0):.2f}")

Leaderboard API

The leaderboard runs as a separate microservice from the agent. It tracks registered wallet addresses and queries Hyperliquid directly for account values.

Deployment

The leaderboard is operated as Nunchi infrastructure. This repository does not publish a public Docker or Railway deployment path for it.

Variable Default Description
LEADERBOARD_PORT 8090 HTTP server port
LEADERBOARD_DB /data/leaderboard.db SQLite database path
CORS_ORIGIN * Allowed CORS origin

GET /health

curl http://localhost:8090/health
{ "status": "ok", "uptime_s": 1200 }

POST /api/register

Register a wallet address for leaderboard tracking. The service queries Hyperliquid at registration time to capture the initial_account_value baseline.

curl -X POST http://localhost:8090/api/register \
  -H "Content-Type: application/json" \
  -d '{
    "address": "0x1234567890abcdef1234567890abcdef12345678",
    "network": "testnet",
    "display_name": "my-agent"
  }'
Param Type Required Description
address string Yes Ethereum address (0x + 40 hex chars)
network string No "testnet" (default) or "mainnet"
display_name string No Display name (max 32 chars)

Response (new registration):

{
  "registered": true,
  "new": true,
  "address": "0x1234567890abcdef1234567890abcdef12345678",
  "network": "testnet",
  "initial_account_value": 10000.0
}

Response (already registered):

{
  "registered": true,
  "new": false,
  "address": "0x1234567890abcdef1234567890abcdef12345678",
  "network": "testnet",
  "initial_account_value": 10000.0
}

Error (invalid address):

{ "error": "Invalid Ethereum address (expected 0x + 40 hex chars)" }

Error (HL query failed):

{ "error": "Failed to query Hyperliquid: ConnectionError(...)" }

How PnL is computed: PnL = current_account_value - initial_account_value. The initial_account_value is captured once at registration and never updated. If the user deposits or withdraws funds after registration, the PnL figure will be incorrect. This is a known limitation.

GET /api/leaderboard

Returns the ranked leaderboard for a given network.

curl "http://localhost:8090/api/leaderboard?network=testnet"
Param Type Default Description
network string testnet "testnet" or "mainnet"

Response:

{
  "agents": [
    {
      "rank": 1,
      "address": "0xabcd...1234",
      "display_name": "alpha-apex",
      "pnl": 842.50,
      "account_value": 10842.50,
      "positions_count": 2,
      "network": "testnet",
      "registered_at": 1709712000000
    },
    {
      "rank": 2,
      "address": "0xef01...5678",
      "display_name": "",
      "pnl": 321.00,
      "account_value": 10321.00,
      "positions_count": 1,
      "network": "testnet",
      "registered_at": 1709798400000
    }
  ],
  "total_agents": 2,
  "last_updated": 1709884800000
}
Field Type Description
agents array Agents sorted by PnL (descending)
agents[].rank int 1-indexed rank
agents[].address string Wallet address (lowercase)
agents[].display_name string User-chosen name (may be empty)
agents[].pnl float current_value - initial_value (rounded to 2 decimals)
agents[].account_value float Current HL account value
agents[].positions_count int Number of open positions on HL
agents[].network string "testnet" or "mainnet"
agents[].registered_at int Registration timestamp (Unix ms)
agents[].stale bool Present and true if the HL query failed for this agent
total_agents int Total registered agents on this network
last_updated int Cache refresh timestamp (Unix ms)

Caching: The leaderboard is cached and refreshed by a background thread every 45 seconds. The last_updated field tells you how fresh the data is. The cache refreshes both testnet and mainnet every cycle.

Frontend polling:

const LEADERBOARD_URL = 'https://leaderboard.nunchi.trade';

async function fetchLeaderboard(network = 'testnet') {
  const res = await fetch(`${LEADERBOARD_URL}/api/leaderboard?network=${network}`);
  const data = await res.json();
  return data.agents;
}

// Poll every 30 seconds
setInterval(() => fetchLeaderboard().then(renderTable), 30000);

CLI usage:

# Register an address
python leaderboard.py register 0x1234...abcd --name "my-agent" --network testnet

# List current rankings
python leaderboard.py list --network testnet

# Start the HTTP server
python leaderboard.py serve --port 8090

MCP Server

The MCP server exposes 16 tools for AI agent orchestration via the Model Context Protocol. This is the access path for Claude Code, OpenClaw, or any MCP-compatible client.

Starting the Server

# stdio transport (for Claude Code / local AI agents)
hl mcp serve

# SSE transport (for development harnesses)
hl mcp serve --transport sse

Hosted users should use the web-auth hosted MCP connection instead of running their own remote MCP deployment.

Connecting from Claude Code

Add to your Claude Code MCP configuration:

{
  "mcpServers": {
    "nunchi": {
      "command": "hl",
      "args": ["mcp", "serve"]
    }
  }
}

Information Tools (Fast, sub-second)

These execute directly in Python with no subprocess overhead.

strategies()

List all 14 trading strategies with descriptions and default parameters.

Tool: strategies
Args: (none)

Returns JSON with strategy catalog and YEX markets. Same data as GET /api/strategies.

builder_status()

Get builder fee configuration.

Tool: builder_status
Args: (none)

Returns:

{
  "enabled": true,
  "builder_address": "0x...",
  "fee_bps": 10,
  "fee_rate_tenths_bps": 100,
  "max_fee_rate_str": "0.1%"
}

wallet_list()

List saved encrypted keystores.

Tool: wallet_list
Args: (none)

Returns array of keystore file paths and addresses, or "No keystores found.".

wallet_auto(save_env=true)

Create a new Ethereum wallet non-interactively. Generates a random private key, encrypts it with a random password, and saves to ~/.hl-agent/keystore/.

Tool: wallet_auto
Args: { "save_env": true }

Returns:

{
  "address": "0x...",
  "password": "random-32-char-token",
  "keystore": "/home/user/.hl-agent/keystore/0x....json",
  "env_file": "/home/user/.hl-agent/env"
}

setup_check()

Validate that the environment is correctly configured for trading.

Tool: setup_check
Args: (none)

Returns:

{
  "ok": [
    "hyperliquid-python-sdk installed",
    "HL_PRIVATE_KEY set",
    "Network: testnet",
    "Builder fee: 10 bps"
  ],
  "issues": [],
  "passed": true
}

Action Tools (Subprocess, seconds to minutes)

These shell out to the CLI and may take significant time.

account(mainnet=false)

Get Hyperliquid account state (balances, margins, positions).

Tool: account
Args: { "mainnet": false }

Returns human-readable account summary from hl account.

status()

Show current positions, PnL, and risk state.

Tool: status
Args: (none)

Returns human-readable status from hl status.

trade(instrument, side, size)

Place a single manual order (IOC).

Tool: trade
Args: { "instrument": "ETH-PERP", "side": "buy", "size": 0.5 }

Executes hl trade ETH-PERP buy 0.5. Returns fill confirmation or error.

run_strategy(strategy, instrument, tick, max_ticks, mock, dry_run, mainnet)

Start autonomous trading with a named strategy.

Tool: run_strategy
Args: {
  "strategy": "avellaneda_mm",
  "instrument": "ETH-PERP",
  "tick": 10,
  "max_ticks": 100,
  "mock": false,
  "dry_run": false,
  "mainnet": false
}

This is a long-running call. If max_ticks is set, it returns after completion. Without max_ticks, it runs indefinitely (set a timeout on your MCP client).

radar_run(mock=false)

Run the opportunity radar once — screen all HL perps for trading setups.

Tool: radar_run
Args: { "mock": false }

Returns scored opportunities with market structure, technicals, and funding analysis.

apex_status()

Get APEX orchestrator status (slots, positions, daily PnL).

Tool: apex_status
Args: (none)

apex_run(mock, max_ticks, preset, mainnet)

Start the APEX multi-slot orchestrator.

Tool: apex_run
Args: {
  "preset": "default",
  "mock": true,
  "max_ticks": 50,
  "mainnet": false
}

Long-running. Timeout is computed as max(120, max_ticks * 60 + 30) seconds.

reflect_run(since=null)

Run REFLECT performance review — analyze trades, compute metrics, generate recommendations.

Tool: reflect_run
Args: { "since": "2026-03-01" }

Returns detailed performance report with win rate, FDR, direction analysis, and parameter adjustment recommendations.

Context Tools (Memory, Journal, Judge)

These access the agent's accumulated knowledge and trade records.

agent_memory(query_type, limit, event_type)

Read agent memory — learnings, parameter changes, market observations.

Tool: agent_memory
Args: { "query_type": "playbook", "limit": 20 }
Arg Values Description
query_type "recent", "playbook" Recent events or accumulated knowledge
limit int Max events (default 20)
event_type "param_change", "reflect_review", "notable_trade", "judge_finding", "session_start", "session_end" Filter by type

trade_journal(date, limit)

Read trade journal — structured position records with entry/exit reasoning.

Tool: trade_journal
Args: { "date": "2026-03-06", "limit": 10 }

Returns journal entries with signal source, entry reasoning, exit reasoning, close reason, and quality rating.

judge_report()

Get latest signal quality evaluation — false positive rates, accuracy by instrument, config recommendations.

Tool: judge_report
Args: (none)

Returns the most recent Judge analysis, or {"status": "no_reports"} if APEX hasn't run long enough to generate one.

obsidian_context()

Read trading context from Obsidian vault — watchlists, market theses, risk preferences.

Tool: obsidian_context
Args: (none)

Requires an Obsidian vault at ~/obsidian-vault. Returns {"status": "unavailable"} if not found.

Reading State Files Directly

If you have filesystem access to the agent (SSH, mounted volume, same container), you can read the state files directly without going through the API.

APEX State

cat $DATA_DIR/apex/state.json | python -m json.tool

Contains: tick_count, slots[], daily_pnl, total_pnl, total_trades, preset, config.

Trade Log

# Last 10 trades
tail -10 $DATA_DIR/apex/trades.jsonl

Each line is a JSON object:

{"tick":42,"oid":"abc123","instrument":"VXX-USDYP","side":"buy","price":"30.55","quantity":"2.5","timestamp_ms":1709712000000,"fee":"0.0345","strategy":"engine_mm"}

Financial values are strings to preserve decimal precision.

StateDB (SQLite)

sqlite3 $DATA_DIR/cli/state.db "SELECT key, value FROM kv"

Keys: tick_count, positions, risk, start_time_ms, strategy_id, instrument, order_stats.

Radar Results

cat $DATA_DIR/radar/scan-history.json | python -m json.tool

Movers Signals

cat $DATA_DIR/movers/scan-history.json | python -m json.tool

REFLECT Reports

ls $DATA_DIR/reflect/

# Reports are saved as markdown files with timestamps
cat $DATA_DIR/reflect/report-latest.md

Journal Entries

tail -5 $DATA_DIR/journal/entries.jsonl

Integration Patterns

Polling Dashboard (Minimal)

import time, requests

AGENT = "https://your-agent.up.railway.app"

while True:
    r = requests.get(f"{AGENT}/api/status")
    data = r.json()

    if data["status"] == "running":
        print(f"Tick {data['tick_count']} | PnL: ${data.get('daily_pnl', 0):.2f}")
        for pos in data.get("positions", []):
            print(f"  {pos['market']} {pos['side']} {pos['size']} @ {pos['entry']} → ROE {pos['roe']:.1f}%")
    else:
        print("Agent stopped")

    time.sleep(30)

SSE Consumer (Real-Time)

import json, requests

AGENT = "https://your-agent.up.railway.app"

with requests.get(f"{AGENT}/api/feed", stream=True) as r:
    for line in r.iter_lines(decode_unicode=True):
        if line and line.startswith("data: "):
            event = json.loads(line[6:])
            tick = event.get("tick_count", 0)
            pnl = event.get("daily_pnl", 0)
            positions = event.get("positions", [])
            print(f"[tick {tick}] PnL=${pnl:.2f} | {len(positions)} positions open")

Multi-Agent Monitor

import requests

agents = [
    {"name": "apex-alpha", "url": "https://apex-alpha.up.railway.app"},
    {"name": "apex-beta", "url": "https://apex-beta.up.railway.app"},
]

for agent in agents:
    try:
        r = requests.get(f"{agent['url']}/api/status", timeout=5)
        data = r.json()
        pnl = data.get("daily_pnl", data.get("realized_pnl", 0))
        print(f"{agent['name']}: {data['status']} | tick {data.get('tick_count', 0)} | PnL ${pnl:.2f}")
    except Exception as e:
        print(f"{agent['name']}: UNREACHABLE ({e})")

Leaderboard + Agent Combo

import requests

AGENT = "https://your-agent.up.railway.app"
LEADERBOARD = "https://leaderboard.nunchi.trade"
YOUR_ADDRESS = "0x1234...abcd"

# Get your agent status
status = requests.get(f"{AGENT}/api/status").json()

# Get leaderboard
lb = requests.get(f"{LEADERBOARD}/api/leaderboard?network=testnet").json()

# Find your rank
your_entry = next((a for a in lb["agents"] if a["address"] == YOUR_ADDRESS.lower()), None)

if your_entry:
    print(f"Rank #{your_entry['rank']} of {lb['total_agents']} | PnL: ${your_entry['pnl']:.2f}")
else:
    print("Not registered on leaderboard")

Endpoint Summary

Agent Endpoints (on every deployed agent)

Method Path Body Response Latency
GET /health None JSON <10ms
GET /api/status None JSON <100ms
GET /api/strategies None JSON <100ms
GET /api/feed None SSE stream Persistent
GET /status None Plain text <1s
POST /api/skill/install None JSON <2s
POST /api/pause Auth header JSON <10ms
POST /api/resume Auth header JSON <10ms

Leaderboard Endpoints (separate service)

Method Path Body Response Latency
GET /health None JSON <10ms
GET /api/leaderboard?network= None JSON <10ms (cached)
POST /api/register JSON JSON 1-5s (queries HL)
OPTIONS * None 204 <10ms

MCP Tools (via hl mcp serve)

Tool Type Latency Side Effects
strategies Fast <100ms None
builder_status Fast <100ms None
wallet_list Fast <100ms None
wallet_auto Fast <500ms Creates keystore file
setup_check Fast <100ms None
account Subprocess 1-5s None
status Subprocess <1s None
trade Subprocess 1-5s Places order on HL
run_strategy Subprocess Minutes+ Runs trading loop
radar_run Subprocess 10-30s None
apex_status Subprocess <1s None
apex_run Subprocess Minutes+ Runs APEX loop
reflect_run Subprocess 5-15s None
agent_memory Fast <100ms None
trade_journal Fast <100ms None
judge_report Fast <100ms None
obsidian_context Fast <100ms None