SolTinel

Multi-agent AI that analyses, guards, and executes Solana trades.
Sentiment. Rug risk. On-chain data. All before the swap.

SolTinel is an open-source trading bot for Solana that runs four specialised AI agents in sequence before touching your wallet. It reads X posts, checks rug risk, analyses on-chain signals, and only executes a trade when every check passes.

Analyst ──▶ Sentiment ──▶ Risk Guard ──approve──▶ Executor
                                      └──reject──▶ (stops, explains why)

New here? Read the Beginner Guide — no prior Solana or AI experience needed.

Why SolTinel?

Most bots execute first and ask questions never. SolTinel flips that:

| What it checks | How | |---|---| | Is the token a rug? | RugCheck.xyz risk score + flagged issues | | What is the market saying? | X posts → LLM sentiment score 0–1 | | Are the on-chain numbers real? | DexScreener: price, liquidity, volume, buy/sell ratio | | Does everything pass? | Hard rules first, then LLM arbitration for edge cases |

If any check fails, the trade is rejected with a plain-English reason. No black box.

Agents

| Agent | Does | |---|---| | Analyst | Fetches price, liquidity, 24h volume, buy/sell ratio via DexScreener (free, no key needed) | | Sentiment | Pulls recent X posts → LLM scores bullish/bearish/neutral with reasoning | | Risk Guard | Combines all signals, applies hard thresholds, asks LLM for grey-zone decisions | | Executor | Jupiter v6 swap — only runs if Risk Guard approves, respects DRY_RUN |

CLI And API

SolTinel now supports both:

• src/cli.ts / soltinel for interactive terminal use
• src/api.ts / package main for embedding in host runtimes like openclaw or hermes

The embeddable API exposes analyzeToken(), runSoltinelSession(), and executeApprovedTrade(), plus host hooks for approval gates and structured progress events.

Quickstart

git clone https://github.com/saadzimat430/soltinel
cd soltinel
cp .env.example .env   # fill in your API keys (see GUIDE.md for help)
npm install
npm run dev            # runs in dry-run mode on BONK by default

To analyse a specific token:

npm run dev -- <SOLANA_TOKEN_MINT_ADDRESS>

No wallet needed to start — DRY_RUN=true by default. You'll see the full agent pipeline run and a detailed rejection or approval reason without any funds at risk.

Configuration (.env)

| Variable | Required | What it does | |---|---|---| | ANTHROPIC_API_KEY | Yes (one of three) | Claude as the LLM brain | | OPENAI_API_KEY | Yes (one of three) | GPT-4o as the LLM brain | | OPENROUTER_API_KEY | Yes (one of three) | Any model via OpenRouter | | SOLANA_RPC_URL | No | Defaults to mainnet-beta public RPC | | SOLANA_PRIVATE_KEY | Only for live trades | Base58 key of your trading wallet | | BIRDEYE_API_KEY | No | Adds holder count to analysis | | X_BEARER_TOKEN | No | Enables real X sentiment (falls back to neutral) | | MAX_SLIPPAGE_BPS | No | Jupiter swap slippage tolerance in basis points (default 100 = 1%) | | RUG_SCORE_MAX | No | Reject if rug risk score exceeds this (0–100, default 40) | | SENTIMENT_MIN | No | Reject if sentiment score is below this (0–1, default 0.35) | | MIN_LIQUIDITY_USD | No | Reject if pool liquidity is below this in USD (default 5000) | | MIN_VOLUME_24H_USD | No | Reject if 24h volume is below this in USD (default 1000) | | MIN_VOL_LIQ_RATIO | No | Reject if volume/liquidity turnover ratio is below this (default 0.01 = 1%) | | MAX_PRICE_CHANGE_24H | No | Reject if 24h price drop exceeds this % (default -25, i.e. reject drops > 25%) | | DRY_RUN | No | true by default — set to false for live trades | | REQUIRE_CONFIRMATION | No | true by default — prompts "y/n" before every real trade; set to false for headless/autonomous mode | | ALLOW_OVERRIDE | No | false by default — when true, shows an override prompt after any rejection so you can proceed at your own risk | | MAX_TRADE_AMOUNT | No | Max input-token amount per trade, defaults to 25 |

Risk policy

Hard rules applied before any LLM call (fail-closed):

• Rug score > RUG_SCORE_MAX → reject
• RugCheck unreachable → reject (never trade blind)
• Sentiment score < SENTIMENT_MIN → reject
• Liquidity < $5k → reject

Borderline cases go to the LLM with full context. Decisions are always typed and logged with a confidence score and plain-English reason.

Extending

Human-in-the-loop — add interruptBefore: ["executorAgent"] to graph.compile({...}) to pause before every trade for manual approval.

Watchlist / cron mode — wrap graph.invoke(...) in a setInterval or BullMQ worker over a list of mints for continuous monitoring.

Persistent state — swap MemorySaver for PostgresSaver from @langchain/langgraph-checkpoint-postgres to survive restarts.

More data sources — add a node for Pump.fun trending, whale wallet tracking, or on-chain order book depth.

Integration API

For agent runtimes and services, import the package instead of shelling out to the CLI:

import { analyzeToken, runSoltinelSession } from "soltinel";

const analysis = await analyzeToken({
  tokenAddress: "DezXAZ8z7PnrnRJjz3wXBoRgixCa6xjnB7YaB1pPB263",
  hooks: {
    onEvent(event) {
      console.log(event.type, event.agent, event.data);
    },
  },
});

const session = await runSoltinelSession({
  tokenAddress: "DezXAZ8z7PnrnRJjz3wXBoRgixCa6xjnB7YaB1pPB263",
  interactionMode: "headless",
  hooks: {
    onEvent(event) {
      console.log(event.type, event.data);
    },
    confirmTrade() {
      return { approved: true, amountUsd: 10 };
    },
    confirmOverride() {
      return false;
    },
    confirmHighPriceImpact() {
      return false;
    },
  },
});

runSoltinelSession() returns machine-readable output including decision, confidence, blockingReasons, recommendedAction, executionStatus, finalAction, and the full typed graph state.

Security

• Never commit .env. Use a dedicated hot wallet funded only to your risk tolerance.
• Keep DRY_RUN=true for at least 10 runs before going live.
• The Risk Guard fails closed — if RugCheck is down, it rejects. Do not remove this.
• Consider running the Executor in a separate process so analysis agents never touch the private key.

Tech stack

TypeScript · LangGraph · LangChain · DexScreener API · RugCheck API · Jupiter v6 API · X API v2 · @solana/web3.js

Contributing

PRs welcome. See CONTRIBUTING.md for guidelines.
Questions? Open an issue or find us on Twitter: @soltinel

License

MIT — free to use, fork, and build on.