LLM Integration

instatollm is designed to produce output that's immediately usable as LLM context. This page shows how to combine it with popular LLMs.

---

Why structured JSON for LLMs

Raw video can't be fed into most LLMs. instatollm converts a Reel into a dense, structured JSON that encodes everything an LLM needs to reason about the content:

• summary — one-sentence context for the LLM
• visual.text_on_screen — OCR output: menus, prices, links, brand names
• audio.transcript — full speech, ready to search or summarize
• content.key_moments — timeline with timestamps
• content.hooks — what grabs attention
• platform_meta — reach, engagement, creator info

---

With Claude (Anthropic)

import anthropic
import httpx
import time

# 1. Get reel analysis
def get_reel_analysis(url: str, ist_token: str) -> dict:
    headers = {"Authorization": f"Bearer {ist_token}"}
    r = httpx.post(
        "https://api.instatollm.com/api/v1/reels/analyze",
        headers=headers,
        json={"url": url},
    )
    job_id = r.json()["id"]

    for _ in range(24):
        time.sleep(5)
        r = httpx.get(f"https://api.instatollm.com/api/v1/reels/{job_id}", headers=headers)
        data = r.json()
        if data["status"] == "done":
            return data["result"]
        if data["status"] == "error":
            raise RuntimeError(data.get("error"))
    raise TimeoutError()


# 2. Pass to Claude
import json

reel = get_reel_analysis(
    "https://www.instagram.com/reel/DYa0IXPgnP5/",
    ist_token="ist_your_token",
)

client = anthropic.Anthropic()
message = client.messages.create(
    model="claude-opus-4-7",
    max_tokens=1024,
    messages=[
        {
            "role": "user",
            "content": f"""Here is a structured analysis of an Instagram Reel:

<reel_analysis>
{json.dumps(reel, indent=2, ensure_ascii=False)}
</reel_analysis>

Based on this analysis:
1. What is the main message of this Reel?
2. What products or places are mentioned (check text_on_screen and transcript)?
3. Who is the target audience?
4. What makes the hook effective?""",
        }
    ],
)
print(message.content[0].text)

---

With OpenAI (GPT-4o)

from openai import OpenAI
import json

reel = get_reel_analysis("https://www.instagram.com/reel/DYa0IXPgnP5/", "ist_your_token")

client = OpenAI()
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[
        {
            "role": "system",
            "content": "You are a social media analyst. Analyze Instagram Reel data and provide actionable insights.",
        },
        {
            "role": "user",
            "content": f"Reel analysis:\n\n{json.dumps(reel, indent=2)}",
        },
    ],
)
print(response.choices[0].message.content)

---

Prompt templates

Content summarization

Given this Instagram Reel analysis:
{reel_json}

Write a 2-sentence summary suitable for a newsletter.
Focus on: what happens, who it's for, key takeaway.

OCR extraction

From this Reel analysis, extract all business-relevant information visible on screen:
- Restaurant/venue names
- Prices and menu items  
- URLs and social handles
- Product names and brands
- Location names

Text on screen: {text_on_screen}
Transcript: {transcript}

Viral analysis

Analyze why this Reel might go viral:

Hook (first 3s): {hooks}
Key moments: {key_moments}
Tone: {tone}
Views: {view_count}
Topics: {topics}

Rate viral potential 1-10 and explain the top 3 reasons.

Content calendar

I'm creating a content calendar. Here's a competitor's successful Reel:

Summary: {summary}
Format: {format}
Target audience: {target_audience}  
Tone: {tone}
Hook: {hooks}
CTA: {call_to_action}

Suggest 3 similar content ideas I could create for my own brand.

---

Batch processing

import asyncio
import httpx
import json

HEADERS = {"Authorization": "Bearer ist_your_token"}

async def analyze_reel(client: httpx.AsyncClient, url: str) -> dict:
    r = await client.post(
        "https://api.instatollm.com/api/v1/reels/analyze",
        json={"url": url},
    )
    job_id = r.json()["id"]

    for _ in range(24):
        await asyncio.sleep(5)
        r = await client.get(f"https://api.instatollm.com/api/v1/reels/{job_id}")
        data = r.json()
        if data["status"] == "done":
            return data["result"]
        if data["status"] == "error":
            raise RuntimeError(data.get("error"))
    raise TimeoutError(f"Timeout for job {job_id}")


async def main():
    urls = [
        "https://www.instagram.com/reel/ABC123/",
        "https://www.instagram.com/reel/DEF456/",
        "https://www.instagram.com/reel/GHI789/",
    ]

    async with httpx.AsyncClient(base_url="https://api.instatollm.com", headers=HEADERS) as client:
        results = await asyncio.gather(*[analyze_reel(client, u) for u in urls])

    for url, result in zip(urls, results):
        print(f"{url}: {result['summary']}")

asyncio.run(main())

---

n8n / Zapier integration

Use the HTTP Request node with:

• Method: POST
• URL: https://api.instatollm.com/api/v1/reels/analyze
• Header: Authorization: Bearer ist_your_token
• Body: {"url": "{{ $json.reel_url }}"}

Then poll GET /api/v1/reels/{{ $json.id }} every 10 seconds until status == "done".