Crypto Sentiment Analysis from X and Reddit with Python

Token Narrative Pulse is a small Python script for crypto sentiment analysis. For a token watchlist it pulls the last 24 hours of X posts and Reddit threads and writes a daily report: mentions per token, unique authors, top posts and a bullish, bearish or neutral breakdown with cited links.

The problem

Crypto narratives move fast on X and Reddit. Analysts want a daily read on which tokens people talk about, how many different people, and in what tone, without bot spam inflating the numbers.

What the app does

• Takes a watchlist of tokens and terms.
• Pulls Top and Latest X posts per term from the last 24 hours.
• Runs AI Search on Reddit and on X plus Reddit for a sentiment breakdown with links.
• Counts mentions and unique authors per token.
• Flags duplicates and template bot posts.
• Writes a report, a CSV of posts and a mentions chart.

How it works

1. X search. GET /twitter through the desearch-py SDK, sort=Top and sort=Latest, start date yesterday.
2. AI Search. POST /desearch/ai/search with tools=["reddit"] and tools=["twitter","reddit"].
3. Clean. Drop posts older than 24 hours, duplicates and template bots.
4. Write the report and a cost log.

The code

async def collect(count, outdir, raw_dir):
    now = dt.datetime.now(dt.timezone.utc)
    since = now - dt.timedelta(hours=24)
    start_date = since.strftime("%Y-%m-%d")  # GET /twitter takes a UTC day, not a timestamp
    costs, x_raw, ai_raw = [], {}, {}
    async with Desearch(api_key=os.environ["DESEARCH_API_KEY"]) as ds:
        print("X search (GET /twitter via desearch-py x_search):")
        for label, (query, _) in WATCHLIST.items():
            for sort in ("Top", "Latest"):
                data = await timed(costs, f"x_search {query!r} sort={sort}",
                                   ds.x_search(query=query, sort=sort, count=count,
                                               start_date=start_date, include_metadata=True))
                x_raw[f"{label}|{sort}"] = dump(data) or []
        print("AI Search (POST /desearch/ai/search via desearch-py ai_search):")
        terms = " OR ".join(q for q, _ in WATCHLIST.values())
        for tools in (["reddit"], ["twitter", "reddit"]):
            data = await timed(costs, f"ai_search tools={tools}",
                               ds.ai_search(prompt=f"What are people saying in the last 24 hours about {terms}?",
                                            tools=tools,
                                            result_type="LINKS_WITH_FINAL_SUMMARY",
                                            system_message=SYSTEM_MESSAGE, count=10,
                                            include_metadata=True))
            ai_raw["+".join(tools)] = dump(data) or {}
    json.dump({"generated_utc": now.isoformat(), "x": x_raw, "ai": ai_raw, "costs": costs},
              open(os.path.join(raw_dir, f"raw-{now:%Y-%m-%d}.json"), "w"), indent=1, default=str)
    return now, since, x_raw, ai_raw, costs

The folder has one script, pulse.py, a sample report in sample-output.md and the chart from that run.

Endpoints

• GET /twitter
• POST /desearch/ai/search

Example output

From a run on 2026-10-03 for a Bittensor watchlist, 24-hour window:

94 posts returned, 85 unique, 63 inside the window. These are counts inside a sample, not total X volume.

Run it yourself

The public examples folder linked above includes the script and a README. Install its requirements, set DESEARCH_API_KEY from console.desearch.ai, and use the run command in that README.

Useful flags: --count (posts per X query, default 20), --from-raw FILE (re-run with no API calls). Edit WATCHLIST in pulse.py to change tokens.

What it costs

One run with the default watchlist made 10 calls, took about 75 seconds and cost $0.026 on 2026-10-03, from the X-Desearch-Cost-Usd headers. See pricing.

Ideas to extend it

• Run it every morning and chart mentions over time.
• Alert when unique authors for a token jump.
• Add on-chain data next to the social view.

{
  "mcpServers": {
    "desearch": {
      "url": "https://mcp.desearch.ai/mcp",
      "headers": {
        "Authorization": "Bearer <DESEARCH_API_KEY>"
      }
    }
  }
}

Use Desearch to pull the last 24 hours of public X posts and Reddit threads for the tokens I name. Summarize mention volume, the strongest posts, and the narrative you can support from those posts. Do not invent prices or posts.