Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.
JavaScript 3d-generationai-agentsanimationavatardigital-humangenerative-aiglbgltfimage-to-3dllmmcpmcp-servermodel-context-protocolsolanatext-to-3dthreejsweb-componentswebglwebxrx402
// agent-sniper — environment + runtime configuration.
//
// Reads and validates the process environment once at boot. Throws on missing
// REQUIRED vars so a misconfigured worker fails loudly instead of trading with
// half a config. Optional vars get sane, conservative defaults.
import { databaseConfigured } from '../../api/_lib/env.js';
function req(name) {
const v = process.env[name];
if (!v || !String(v).trim()) {
throw new Error(`[agent-sniper] missing required env var: ${name}`);
}
return v;
}
// The Postgres connection string may arrive under DATABASE_URL or any standard
// Vercel-Postgres/Neon alias; db.js resolves the live connection from that set,
// so assert configuration the same way rather than hard-requiring the bare name.
function requireDatabase(scope) {
if (!databaseConfigured()) {
throw new Error(`[${scope}] missing required env var: DATABASE_URL (or a POSTGRES_URL/NEON_DATABASE_URL alias)`);
}
}
function num(name, def) {
const raw = process.env[name];
if (raw == null || raw === '') return def;
const n = Number(raw);
return Number.isFinite(n) ? n : def;
}
function bool(name, def = false) {
const raw = process.env[name];
if (raw == null || raw === '') return def;
return /^(1|true|yes|on)$/i.test(String(raw).trim());
}
// Parse a comma/space/newline-separated allowlist of agent UUIDs. Returns a
// de-duped array, or null when unset/empty (null = "all agents", the default).
// Exported for unit testing — the scoping fix that keeps a worker from acting on
// every armed strategy in the shared DB depends on this being exact.
const UUID_RE = /^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/i;
export function parseAgentIds(raw) {
if (raw == null || String(raw).trim() === '') return null;
const ids = String(raw)
.split(/[\s,]+/)
.map((s) => s.trim().toLowerCase())
.filter(Boolean)
.filter((s) => UUID_RE.test(s));
if (!ids.length) return null;
return [...new Set(ids)];
}
// Resolve the RPC endpoint the worker's read paths (e.g. the Mayhem curve read)
// should use. Prefer an explicit URL, then a Helius key, else empty (callers
// fall back to a public mainnet endpoint).
function resolveRpcUrl() {
if (process.env.SOLANA_RPC_URL) return process.env.SOLANA_RPC_URL.trim();
if (process.env.HELIUS_API_KEY) return `https://mainnet.helius-rpc.com/?api-key=${process.env.HELIUS_API_KEY.trim()}`;
return '';
}
export function loadConfig() {
// DATABASE_URL + JWT_SECRET are consumed transitively by db.js / agent-wallet.js.
// We assert them here so the failure points at config, not a cryptic decrypt
// error three layers deep on the first trade.
requireDatabase('agent-sniper');
req('JWT_SECRET');
const network = (process.env.SNIPER_NETWORK || 'mainnet').trim();
if (network !== 'mainnet' && network !== 'devnet') {
throw new Error(`[agent-sniper] SNIPER_NETWORK must be mainnet|devnet, got "${network}"`);
}
const mode = (process.env.SNIPER_MODE || 'simulate').trim();
if (mode !== 'live' && mode !== 'simulate') {
throw new Error(`[agent-sniper] SNIPER_MODE must be live|simulate, got "${mode}"`);
}
// Live trading on a public RPC will 429 under the new-mint firehose. Refuse to
// start live without a real endpoint rather than silently dropping trades.
if (mode === 'live' && !process.env.SOLANA_RPC_URL && !process.env.HELIUS_API_KEY) {
throw new Error(
'[agent-sniper] live mode requires SOLANA_RPC_URL or HELIUS_API_KEY (public RPC will rate-limit)',
);
}
return {
network,
mode,
// Resolved RPC endpoint, shared by the worker's read paths (Mayhem curve read).
rpcUrl: resolveRpcUrl(),
// ── agent scoping ────────────────────────────────────────────────────────
// Optional allowlist of agent UUIDs. When set, the worker acts ONLY on these
// agents' strategies — the rest of the shared DB's armed strategies are
// ignored. Unset (null) = every enabled strategy for the network, the legacy
// behaviour. Set this to run a bounded experiment/fleet against the prod DB
// without entangling it with every other armed agent.
agentIds: parseAgentIds(process.env.SNIPER_AGENT_IDS),
// ── Mayhem exclusion (owner rule) ────────────────────────────────────────
// NEVER buy pump.fun "Mayhem"-mode tokens — only regular launches. Reads the
// bonding curve's isMayhemMode once per mint (cached). On by default; set
// SNIPER_MAYHEM_FILTER=0 to disable. FAIL CLOSED by default: when the curve
// can't be read we skip the buy, because we can't confirm it isn't Mayhem —
// and the owner rule is NEVER buy Mayhem, not "buy when unsure". A fresh
// mint's curve exists at creation, so an unreadable read is an RPC hiccup, not
// a missing account; the right response is to pass on that coin, not buy blind.
// Set SNIPER_MAYHEM_STRICT=0 to restore allow-on-unknown (raw speed).
mayhemFilter: bool('SNIPER_MAYHEM_FILTER', true),
mayhemStrict: bool('SNIPER_MAYHEM_STRICT', true),
// Global emergency stop — set SNIPER_GLOBAL_KILL=1 to halt all new buys
// while still letting the position loop manage/exit open positions.
globalKill: bool('SNIPER_GLOBAL_KILL', false),
// ── adversarial Risk Officer ─────────────────────────────────────────────
// Fleet-wide DEFAULT enforcement level for the independent pre-trade review
// (workers/agent-sniper/risk-officer.js); a strategy's own
// `risk_officer_level` column overrides it. 'shadow' (the default) records
// the veto it would have cast and changes nothing; 'enforce' lets a veto
// abort the buy and a smaller suggested size shrink it; 'off' skips it.
// Deliberately NOT 'enforce' out of the box: enforcement changes what the
// live fleet buys with real SOL, so arming it is an owner decision made
// against the evidence in sniper_risk_reviews.
riskOfficer: (process.env.SNIPER_RISK_OFFICER || 'shadow').trim().toLowerCase(),
// Position re-quote / exit-evaluation cadence.
pollMs: Math.max(1_000, num('SNIPER_POLL_MS', 5_000)),
// Liquidity-decay exit: an underwater position whose quoted value has not
// moved AT ALL for this long (no one is trading the coin) exits early with
// reason 'liquidity_decay' instead of squatting on a concurrency slot until
// the 30-minute timeout. 0 disables. Default 5 minutes: on a minutes-old
// pump.fun launch, five minutes of literally zero trades is a dead market.
liquidityDecayS: Math.max(0, num('SNIPER_LIQUIDITY_DECAY_S', 300)),
// How often the strategy cache is refreshed from the DB.
strategyRefreshMs: Math.max(5_000, num('SNIPER_STRATEGY_REFRESH_MS', 15_000)),
// Platform-wide buy throttle — a backstop independent of per-agent caps.
maxGlobalBuysPerMin: Math.max(0, num('SNIPER_MAX_GLOBAL_BUYS_PER_MIN', 10)),
// Watchdog: if the feed delivers nothing for this long, re-subscribe.
feedWatchdogMs: Math.max(30_000, num('SNIPER_FEED_WATCHDOG_MS', 180_000)),
// Confirmation timeout for a broadcast trade.
confirmTimeoutMs: Math.max(15_000, num('SNIPER_CONFIRM_TIMEOUT_MS', 60_000)),
// Owner directive: the fleet is a learning experiment, not a profit desk — a
// wallet too poor for its strategy's configured per_trade_lamports should
// shrink the trade to whatever it actually has left rather than sit idle.
// This is the floor below that shrink: once even a dust-sized buy wouldn't
// clear it, THEN skip with insufficient_sol. Default 0.00001 SOL.
minTradeLamports: BigInt(Math.max(1, Math.round(num('SNIPER_MIN_TRADE_LAMPORTS', 10_000)))),
// ── liveness + alerting ──────────────────────────────────────────────────
// How often the worker upserts its bot_heartbeat row. /api/sniper/status
// (and /status) read this to answer "is the sniper alive?" without SSH.
heartbeatMs: Math.max(10_000, num('SNIPER_HEARTBEAT_MS', 30_000)),
// Executor/RPC errors within this sliding window that trip an ops alert.
errorAlertThreshold: Math.max(1, num('SNIPER_ERROR_ALERT_THRESHOLD', 5)),
errorAlertWindowMs: Math.max(60_000, num('SNIPER_ERROR_ALERT_WINDOW_MS', 600_000)),
// Announce boot/shutdown to the ops channel (a deploy that never comes back
// up is otherwise silent). Disable in noisy dev with SNIPER_ANNOUNCE=0.
announceLifecycle: bool('SNIPER_ANNOUNCE', true),
// ── first-claim trigger ─────────────────────────────────────────────────
// How often to poll the on-chain fee-claim stream for first-ever claims.
claimPollMs: Math.max(10_000, num('SNIPER_CLAIM_POLL_MS', 30_000)),
// Window scanned each poll — must comfortably exceed the poll interval so a
// claim can't slip between scans.
claimLookbackSeconds: Math.max(60, num('SNIPER_CLAIM_LOOKBACK_S', 600)),
// Default freshness gate: ignore a first claim older than this when first
// observed (a per-strategy first_claim_max_age_seconds overrides it). Keeps
// a worker restart / backfill from sniping a stale claim.
claimMaxAgeSeconds: Math.max(30, num('SNIPER_CLAIM_MAX_AGE_S', 300)),
// ── coin intelligence engine ────────────────────────────────────────────
// Observe every new coin's first seconds of trading, classify it, persist
// signals, and drive intel_confirmed strategies. On by default — it's the
// platform's eyes. Set SNIPER_INTEL=0 to disable (e.g. a trade-only worker).
intel: bool('SNIPER_INTEL', true),
// Observation window per coin before signals are finalized (ms).
intelWindowMs: Math.max(15_000, num('SNIPER_INTEL_WINDOW_MS', 90_000)),
// Max coins observed simultaneously — a backstop on memory + WS subs.
intelMaxConcurrent: Math.max(50, num('SNIPER_INTEL_MAX_CONCURRENT', 400)),
// Refine classification with the LLM (free-first). Off by default: the
// deterministic heuristic is instant and free; LLM adds ~1 call per coin.
intelLlm: bool('SNIPER_INTEL_LLM', false),
// ── pre-launch creator-wallet radar (task 04) ───────────────────────────
// Watch proven creator + smart-money wallets on-chain and detect launch
// precursors (funding a fresh deploy wallet / a pump create) at block-0 to
// pre-arm a snipe on signal rather than luck. On by default — it's the
// platform's earliest edge. SNIPER_RADAR=0 disables it (degrades to the
// normal feed-based snipe, reported honestly as paused).
radar: bool('SNIPER_RADAR', true),
// How often each watched wallet's recent signatures are polled.
radarPollMs: Math.max(5_000, num('SNIPER_RADAR_POLL_MS', 15_000)),
// Watched wallets scanned per tick (bounds RPC pressure; the set is covered
// round-robin across ticks).
radarWalletsPerTick: Math.max(5, num('SNIPER_RADAR_WALLETS_PER_TICK', 60)),
// How often the watchlist is auto-recurated from the graph.
radarWatchlistRefreshMs: Math.max(60_000, num('SNIPER_RADAR_WATCHLIST_REFRESH_MS', 300_000)),
// Creator pedigree floor for auto-inclusion (graduated coins).
radarMinCreatorGraduated: Math.max(1, num('SNIPER_RADAR_MIN_GRADUATED', 2)),
// Reputation floor for smart-money auto-inclusion (0..100).
radarSmartMoneyMinScore: Math.max(0, Math.min(100, num('SNIPER_RADAR_SM_MIN_SCORE', 70))),
// Hard cap on the monitored set (stale low-signal wallets are evicted).
radarMaxWatch: Math.max(20, num('SNIPER_RADAR_MAX_WATCH', 500)),
// Default freshness gate for a precursor before it can pre-arm (ms). A
// per-strategy radar_max_age_ms overrides it.
radarMaxAgeMs: Math.max(10_000, num('SNIPER_RADAR_MAX_AGE_MS', 120_000)),
// A SOL outflow at least this large to a fresh wallet counts as deploy
// funding (rent + first buy). Below it is ignored as noise.
radarMinFundingLamports: Math.max(0, num('SNIPER_RADAR_MIN_FUNDING_LAMPORTS', 30_000_000)),
// How long a freshly-funded wallet stays under correlation watch for its
// create instruction before it's dropped (ms).
radarDeployWatchTtlMs: Math.max(30_000, num('SNIPER_RADAR_DEPLOY_WATCH_TTL_MS', 180_000)),
// ── autonomous coin launcher ────────────────────────────────────────────
// Runs a periodic check for enabled launcher configs and fires launches.
// On by default when AGENT_JWT is set; disable with SNIPER_LAUNCHER=0.
launcher: bool('SNIPER_LAUNCHER', !!process.env.AGENT_JWT),
// How often to check for due launcher configs (ms).
launcherPollMs: Math.max(30_000, num('SNIPER_LAUNCHER_POLL_MS', 60_000)),
// ── sentiment-flip exit ─────────────────────────────────────────────────
// Let the per-coin x402 sentiment the worker already pays for drive a sell:
// cut an underwater position early when its signal flips confidently bearish,
// ahead of the hard stop-loss. Off by default — existing strategies keep their
// exact stop/trailing/TP behaviour until an operator opts in.
exitOnBearish: bool('SNIPER_EXIT_ON_BEARISH', false),
// Minimum sentiment confidence (0..1) before a bearish flip can exit.
exitBearishMinConfidence: Math.max(0, Math.min(1, num('SNIPER_EXIT_BEARISH_MIN_CONFIDENCE', 0.7))),
// ── buy-side auto-funding ───────────────────────────────────────────────
// Keeps each armed agent's own wallet topped up from the launcher master so
// a hot sniper never silently runs dry and starts failing every buy. On by
// default; disable with SNIPER_AUTO_FUND=0. Per-transfer / daily caps and
// the floor/target are read in auto-funder.js (SNIPER_AUTO_FUND_*).
autoFund: bool('SNIPER_AUTO_FUND', true),
// ── creator auto-claim ──────────────────────────────────────────────────
// Polls claimable creator fees on agent-launched coins and auto-claims them.
// On by default when AGENT_JWT is set; disable with SNIPER_AUTO_CLAIM=0.
autoClaim: bool('SNIPER_AUTO_CLAIM', !!process.env.AGENT_JWT),
// How often to poll claimable fees (ms). Default 5 minutes.
autoClaimPollMs: Math.max(60_000, num('SNIPER_AUTO_CLAIM_POLL_MS', 300_000)),
// ── market maker ───────────────────────────────────────────────────────
// Runs range-based market making on configured coins via Jito execution.
// On by default; disable with SNIPER_MARKET_MAKER=0.
marketMaker: bool('SNIPER_MARKET_MAKER', true),
// Default rebalance interval when no per-config override (ms).
marketMakerIntervalMs: Math.max(5_000, num('SNIPER_MM_INTERVAL_MS', 10_000)),
};
}