GPT-5 Nano

import { createClient } from '@runware/sdk'

const client = await createClient({ apiKey: process.env.RUNWARE_API_KEY })
await client.connect()

const [result] = await client.run({
  model: 'openai:gpt@5-nano',
  settings: {
    systemPrompt: 'You are a podcast editorial producer. Turn source transcripts into accurate, engaging release materials without inventing facts, quotes, timestamps, or credentials. Preserve scientific nuance, explain technical terms plainly, and distinguish preliminary findings from established evidence. Return clean, publication-ready copy with the requested headings.',
    maxTokens: 2200,
    thinkingLevel: 'low'
  },
  messages: [
    {
      role: 'user',
      content: 'Create the launch kit for an episode of Signal & Soil, a practical climate-science podcast for informed general audiences.\n\nDELIVERABLES\n1. Five episode-title options, each under 65 characters and free of clickbait.\n2. A 90–120 word episode summary.\n3. Six chapter titles paired with the approximate timestamps supplied in the transcript.\n4. Three key takeaways, each one sentence.\n5. A LinkedIn post of 100–140 words for the podcast account.\n6. Two social captions under 220 characters each: one curious, one pragmatic.\n7. A factual accuracy note listing any claims the editor should verify before publication.\n\nSTYLE\nClear, intelligent, warm, and grounded. Emphasize practical implications for farmers and food systems. Do not imply that biochar is a universal solution. Use only information in the transcript. Paraphrase rather than fabricating direct quotes.\n\nSOURCE TRANSCRIPT\n[00:00] HOST MAYA CHEN: Today we are asking whether biochar can help farms hold more water while storing carbon. Our guest is Dr. Lena Ortiz, a soil biogeochemist at Great Plains State University. Her team has completed the second growing season of a three-year field trial in western Kansas.\n\n[02:10] ORTIZ: Biochar is a carbon-rich material made by heating crop residues in a low-oxygen chamber. The process is called pyrolysis. The properties of the final material depend heavily on the feedstock and temperature, so biochar is not one uniform product.\n\n[05:35] ORTIZ: In our trial, we incorporated locally produced wheat-straw biochar into 24 dryland sorghum plots. We are comparing two application rates with untreated control plots. These are interim observations, not final results.\n\n[08:20] ORTIZ: During the first season, soil moisture in the upper 20 centimeters was modestly higher in the amended plots after light rain. We did not see a statistically significant yield difference. In the second season, a severe heat wave complicated interpretation, and the laboratory analysis is still underway.\n\n[12:45] HOST: Farmers often ask about cost. ORTIZ: That is currently the largest barrier. The biochar itself, transport, and field application can be expensive. It makes more sense to investigate regional systems using nearby residues than to haul bulky material long distances.\n\n[16:30] ORTIZ: People sometimes describe biochar as permanent carbon storage. That is too simple. A resistant fraction can remain in soil for a long time, but persistence varies with production conditions, soil type, and climate. Our project is measuring changes over three years; it cannot directly prove century-scale storage.\n\n[20:05] HOST: Are there risks? ORTIZ: Poorly characterized material may contain contaminants or alter soil pH in unhelpful ways. Producers should request an analysis of the material and begin with small trials. Biochar also should not replace erosion control, crop rotation, or careful nutrient management.\n\n[24:15] ORTIZ: The most promising model may connect waste management with soil improvement. A cooperative could convert clean local residues, use process heat, test every batch, and apply the material nearby. But the economics and emissions of the entire system need to be counted.\n\n[28:40] HOST: What should listeners remember? ORTIZ: Biochar is a tool with interesting potential, not a shortcut. Match the material to the soil, verify quality, and look at the full system from residue source to field performance.\n\n[30:10] END'
    }
  ]
})

import asyncio
import os

from runware import Runware


async def main():
    async with Runware(api_key=os.environ["RUNWARE_API_KEY"]) as client:
        results = await client.run({
            "model": "openai:gpt@5-nano",
            "settings": {
                "systemPrompt": "You are a podcast editorial producer. Turn source transcripts into accurate, engaging release materials without inventing facts, quotes, timestamps, or credentials. Preserve scientific nuance, explain technical terms plainly, and distinguish preliminary findings from established evidence. Return clean, publication-ready copy with the requested headings.",
                "maxTokens": 2200,
                "thinkingLevel": "low"
            },
            "messages": [
                {
                    "role": "user",
                    "content": "Create the launch kit for an episode of Signal & Soil, a practical climate-science podcast for informed general audiences.\n\nDELIVERABLES\n1. Five episode-title options, each under 65 characters and free of clickbait.\n2. A 90–120 word episode summary.\n3. Six chapter titles paired with the approximate timestamps supplied in the transcript.\n4. Three key takeaways, each one sentence.\n5. A LinkedIn post of 100–140 words for the podcast account.\n6. Two social captions under 220 characters each: one curious, one pragmatic.\n7. A factual accuracy note listing any claims the editor should verify before publication.\n\nSTYLE\nClear, intelligent, warm, and grounded. Emphasize practical implications for farmers and food systems. Do not imply that biochar is a universal solution. Use only information in the transcript. Paraphrase rather than fabricating direct quotes.\n\nSOURCE TRANSCRIPT\n[00:00] HOST MAYA CHEN: Today we are asking whether biochar can help farms hold more water while storing carbon. Our guest is Dr. Lena Ortiz, a soil biogeochemist at Great Plains State University. Her team has completed the second growing season of a three-year field trial in western Kansas.\n\n[02:10] ORTIZ: Biochar is a carbon-rich material made by heating crop residues in a low-oxygen chamber. The process is called pyrolysis. The properties of the final material depend heavily on the feedstock and temperature, so biochar is not one uniform product.\n\n[05:35] ORTIZ: In our trial, we incorporated locally produced wheat-straw biochar into 24 dryland sorghum plots. We are comparing two application rates with untreated control plots. These are interim observations, not final results.\n\n[08:20] ORTIZ: During the first season, soil moisture in the upper 20 centimeters was modestly higher in the amended plots after light rain. We did not see a statistically significant yield difference. In the second season, a severe heat wave complicated interpretation, and the laboratory analysis is still underway.\n\n[12:45] HOST: Farmers often ask about cost. ORTIZ: That is currently the largest barrier. The biochar itself, transport, and field application can be expensive. It makes more sense to investigate regional systems using nearby residues than to haul bulky material long distances.\n\n[16:30] ORTIZ: People sometimes describe biochar as permanent carbon storage. That is too simple. A resistant fraction can remain in soil for a long time, but persistence varies with production conditions, soil type, and climate. Our project is measuring changes over three years; it cannot directly prove century-scale storage.\n\n[20:05] HOST: Are there risks? ORTIZ: Poorly characterized material may contain contaminants or alter soil pH in unhelpful ways. Producers should request an analysis of the material and begin with small trials. Biochar also should not replace erosion control, crop rotation, or careful nutrient management.\n\n[24:15] ORTIZ: The most promising model may connect waste management with soil improvement. A cooperative could convert clean local residues, use process heat, test every batch, and apply the material nearby. But the economics and emissions of the entire system need to be counted.\n\n[28:40] HOST: What should listeners remember? ORTIZ: Biochar is a tool with interesting potential, not a shortcut. Match the material to the soil, verify quality, and look at the full system from residue source to field performance.\n\n[30:10] END"
                }
            ]
        })


asyncio.run(main())

curl https://api.runware.ai/v1 \
  -H "Authorization: Bearer $RUNWARE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '[
    {
      "taskType": "textInference",
      "taskUUID": "fb733ee2-f10b-48a6-a9f2-69e8682777e9",
      "model": "openai:gpt@5-nano",
      "settings": {
        "systemPrompt": "You are a podcast editorial producer. Turn source transcripts into accurate, engaging release materials without inventing facts, quotes, timestamps, or credentials. Preserve scientific nuance, explain technical terms plainly, and distinguish preliminary findings from established evidence. Return clean, publication-ready copy with the requested headings.",
        "maxTokens": 2200,
        "thinkingLevel": "low"
      },
      "messages": [
        {
          "role": "user",
          "content": "Create the launch kit for an episode of Signal & Soil, a practical climate-science podcast for informed general audiences.\n\nDELIVERABLES\n1. Five episode-title options, each under 65 characters and free of clickbait.\n2. A 90–120 word episode summary.\n3. Six chapter titles paired with the approximate timestamps supplied in the transcript.\n4. Three key takeaways, each one sentence.\n5. A LinkedIn post of 100–140 words for the podcast account.\n6. Two social captions under 220 characters each: one curious, one pragmatic.\n7. A factual accuracy note listing any claims the editor should verify before publication.\n\nSTYLE\nClear, intelligent, warm, and grounded. Emphasize practical implications for farmers and food systems. Do not imply that biochar is a universal solution. Use only information in the transcript. Paraphrase rather than fabricating direct quotes.\n\nSOURCE TRANSCRIPT\n[00:00] HOST MAYA CHEN: Today we are asking whether biochar can help farms hold more water while storing carbon. Our guest is Dr. Lena Ortiz, a soil biogeochemist at Great Plains State University. Her team has completed the second growing season of a three-year field trial in western Kansas.\n\n[02:10] ORTIZ: Biochar is a carbon-rich material made by heating crop residues in a low-oxygen chamber. The process is called pyrolysis. The properties of the final material depend heavily on the feedstock and temperature, so biochar is not one uniform product.\n\n[05:35] ORTIZ: In our trial, we incorporated locally produced wheat-straw biochar into 24 dryland sorghum plots. We are comparing two application rates with untreated control plots. These are interim observations, not final results.\n\n[08:20] ORTIZ: During the first season, soil moisture in the upper 20 centimeters was modestly higher in the amended plots after light rain. We did not see a statistically significant yield difference. In the second season, a severe heat wave complicated interpretation, and the laboratory analysis is still underway.\n\n[12:45] HOST: Farmers often ask about cost. ORTIZ: That is currently the largest barrier. The biochar itself, transport, and field application can be expensive. It makes more sense to investigate regional systems using nearby residues than to haul bulky material long distances.\n\n[16:30] ORTIZ: People sometimes describe biochar as permanent carbon storage. That is too simple. A resistant fraction can remain in soil for a long time, but persistence varies with production conditions, soil type, and climate. Our project is measuring changes over three years; it cannot directly prove century-scale storage.\n\n[20:05] HOST: Are there risks? ORTIZ: Poorly characterized material may contain contaminants or alter soil pH in unhelpful ways. Producers should request an analysis of the material and begin with small trials. Biochar also should not replace erosion control, crop rotation, or careful nutrient management.\n\n[24:15] ORTIZ: The most promising model may connect waste management with soil improvement. A cooperative could convert clean local residues, use process heat, test every batch, and apply the material nearby. But the economics and emissions of the entire system need to be counted.\n\n[28:40] HOST: What should listeners remember? ORTIZ: Biochar is a tool with interesting potential, not a shortcut. Match the material to the soil, verify quality, and look at the full system from residue source to field performance.\n\n[30:10] END"
        }
      ]
    }
  ]'

runware run openai:gpt@5-nano \
  settings.systemPrompt="You are a podcast editorial producer. Turn source transcripts into accurate, engaging release materials without inventing facts, quotes, timestamps, or credentials. Preserve scientific nuance, explain technical terms plainly, and distinguish preliminary findings from established evidence. Return clean, publication-ready copy with the requested headings." \
  settings.maxTokens=2200 \
  settings.thinkingLevel=low \
  messages.0.role=user \
  messages.0.content="Create the launch kit for an episode of Signal & Soil, a practical climate-science podcast for informed general audiences.

DELIVERABLES
1. Five episode-title options, each under 65 characters and free of clickbait.
2. A 90–120 word episode summary.
3. Six chapter titles paired with the approximate timestamps supplied in the transcript.
4. Three key takeaways, each one sentence.
5. A LinkedIn post of 100–140 words for the podcast account.
6. Two social captions under 220 characters each: one curious, one pragmatic.
7. A factual accuracy note listing any claims the editor should verify before publication.

STYLE
Clear, intelligent, warm, and grounded. Emphasize practical implications for farmers and food systems. Do not imply that biochar is a universal solution. Use only information in the transcript. Paraphrase rather than fabricating direct quotes.

SOURCE TRANSCRIPT
[00:00] HOST MAYA CHEN: Today we are asking whether biochar can help farms hold more water while storing carbon. Our guest is Dr. Lena Ortiz, a soil biogeochemist at Great Plains State University. Her team has completed the second growing season of a three-year field trial in western Kansas.

[02:10] ORTIZ: Biochar is a carbon-rich material made by heating crop residues in a low-oxygen chamber. The process is called pyrolysis. The properties of the final material depend heavily on the feedstock and temperature, so biochar is not one uniform product.

[05:35] ORTIZ: In our trial, we incorporated locally produced wheat-straw biochar into 24 dryland sorghum plots. We are comparing two application rates with untreated control plots. These are interim observations, not final results.

[08:20] ORTIZ: During the first season, soil moisture in the upper 20 centimeters was modestly higher in the amended plots after light rain. We did not see a statistically significant yield difference. In the second season, a severe heat wave complicated interpretation, and the laboratory analysis is still underway.

[12:45] HOST: Farmers often ask about cost. ORTIZ: That is currently the largest barrier. The biochar itself, transport, and field application can be expensive. It makes more sense to investigate regional systems using nearby residues than to haul bulky material long distances.

[16:30] ORTIZ: People sometimes describe biochar as permanent carbon storage. That is too simple. A resistant fraction can remain in soil for a long time, but persistence varies with production conditions, soil type, and climate. Our project is measuring changes over three years; it cannot directly prove century-scale storage.

[20:05] HOST: Are there risks? ORTIZ: Poorly characterized material may contain contaminants or alter soil pH in unhelpful ways. Producers should request an analysis of the material and begin with small trials. Biochar also should not replace erosion control, crop rotation, or careful nutrient management.

[24:15] ORTIZ: The most promising model may connect waste management with soil improvement. A cooperative could convert clean local residues, use process heat, test every batch, and apply the material nearby. But the economics and emissions of the entire system need to be counted.

[28:40] HOST: What should listeners remember? ORTIZ: Biochar is a tool with interesting potential, not a shortcut. Match the material to the soil, verify quality, and look at the full system from residue source to field performance.

[30:10] END"

{
  "taskType": "textInference",
  "taskUUID": "fb733ee2-f10b-48a6-a9f2-69e8682777e9",
  "model": "openai:gpt@5-nano",
  "settings": {
    "systemPrompt": "You are a podcast editorial producer. Turn source transcripts into accurate, engaging release materials without inventing facts, quotes, timestamps, or credentials. Preserve scientific nuance, explain technical terms plainly, and distinguish preliminary findings from established evidence. Return clean, publication-ready copy with the requested headings.",
    "maxTokens": 2200,
    "thinkingLevel": "low"
  },
  "messages": [
    {
      "role": "user",
      "content": "Create the launch kit for an episode of Signal & Soil, a practical climate-science podcast for informed general audiences.\n\nDELIVERABLES\n1. Five episode-title options, each under 65 characters and free of clickbait.\n2. A 90–120 word episode summary.\n3. Six chapter titles paired with the approximate timestamps supplied in the transcript.\n4. Three key takeaways, each one sentence.\n5. A LinkedIn post of 100–140 words for the podcast account.\n6. Two social captions under 220 characters each: one curious, one pragmatic.\n7. A factual accuracy note listing any claims the editor should verify before publication.\n\nSTYLE\nClear, intelligent, warm, and grounded. Emphasize practical implications for farmers and food systems. Do not imply that biochar is a universal solution. Use only information in the transcript. Paraphrase rather than fabricating direct quotes.\n\nSOURCE TRANSCRIPT\n[00:00] HOST MAYA CHEN: Today we are asking whether biochar can help farms hold more water while storing carbon. Our guest is Dr. Lena Ortiz, a soil biogeochemist at Great Plains State University. Her team has completed the second growing season of a three-year field trial in western Kansas.\n\n[02:10] ORTIZ: Biochar is a carbon-rich material made by heating crop residues in a low-oxygen chamber. The process is called pyrolysis. The properties of the final material depend heavily on the feedstock and temperature, so biochar is not one uniform product.\n\n[05:35] ORTIZ: In our trial, we incorporated locally produced wheat-straw biochar into 24 dryland sorghum plots. We are comparing two application rates with untreated control plots. These are interim observations, not final results.\n\n[08:20] ORTIZ: During the first season, soil moisture in the upper 20 centimeters was modestly higher in the amended plots after light rain. We did not see a statistically significant yield difference. In the second season, a severe heat wave complicated interpretation, and the laboratory analysis is still underway.\n\n[12:45] HOST: Farmers often ask about cost. ORTIZ: That is currently the largest barrier. The biochar itself, transport, and field application can be expensive. It makes more sense to investigate regional systems using nearby residues than to haul bulky material long distances.\n\n[16:30] ORTIZ: People sometimes describe biochar as permanent carbon storage. That is too simple. A resistant fraction can remain in soil for a long time, but persistence varies with production conditions, soil type, and climate. Our project is measuring changes over three years; it cannot directly prove century-scale storage.\n\n[20:05] HOST: Are there risks? ORTIZ: Poorly characterized material may contain contaminants or alter soil pH in unhelpful ways. Producers should request an analysis of the material and begin with small trials. Biochar also should not replace erosion control, crop rotation, or careful nutrient management.\n\n[24:15] ORTIZ: The most promising model may connect waste management with soil improvement. A cooperative could convert clean local residues, use process heat, test every batch, and apply the material nearby. But the economics and emissions of the entire system need to be counted.\n\n[28:40] HOST: What should listeners remember? ORTIZ: Biochar is a tool with interesting potential, not a shortcut. Match the material to the soil, verify quality, and look at the full system from residue source to field performance.\n\n[30:10] END"
    }
  ]
}

import { createClient } from '@runware/sdk'

const client = await createClient({ apiKey: process.env.RUNWARE_API_KEY })
await client.connect()

const [result] = await client.run({
  model: 'openai:gpt@5-nano',
  settings: {
    systemPrompt: 'You are an experienced warehouse EHS inspection assistant. Analyze only conditions that are directly visible in the supplied image. Do not infer measurements, legal violations, worker intent, or facts outside the frame. Produce a concise professional report with: an executive summary; a numbered findings table containing severity, visible evidence, hazard category, potential consequence, and recommended immediate action; positive safety observations; and items requiring an on-site inspection. Assign severity as Critical, High, Medium, or Low and explain the visual basis. State that the report is a preliminary visual screening and not a substitute for an inspection by a qualified safety professional.',
    maxTokens: 2500,
    thinkingLevel: 'low'
  },
  inputs: {
    images: [
      'https://assets.runware.ai/assets/inputs/510aa7c7-4108-4de0-ba53-5565a164a085.jpg'
    ]
  },
  messages: [
    {
      role: 'user',
      content: 'Review the attached photograph from Receiving Bay 4 at a regional ecommerce distribution center. Prepare a shift-ready safety inspection report for the site EHS manager. Prioritize hazards that need immediate isolation or correction, cite where each condition appears in the image, and avoid reporting anything that is not visibly supported.'
    }
  ]
})

import asyncio
import os

from runware import Runware


async def main():
    async with Runware(api_key=os.environ["RUNWARE_API_KEY"]) as client:
        results = await client.run({
            "model": "openai:gpt@5-nano",
            "settings": {
                "systemPrompt": "You are an experienced warehouse EHS inspection assistant. Analyze only conditions that are directly visible in the supplied image. Do not infer measurements, legal violations, worker intent, or facts outside the frame. Produce a concise professional report with: an executive summary; a numbered findings table containing severity, visible evidence, hazard category, potential consequence, and recommended immediate action; positive safety observations; and items requiring an on-site inspection. Assign severity as Critical, High, Medium, or Low and explain the visual basis. State that the report is a preliminary visual screening and not a substitute for an inspection by a qualified safety professional.",
                "maxTokens": 2500,
                "thinkingLevel": "low"
            },
            "inputs": {
                "images": [
                    "https://assets.runware.ai/assets/inputs/510aa7c7-4108-4de0-ba53-5565a164a085.jpg"
                ]
            },
            "messages": [
                {
                    "role": "user",
                    "content": "Review the attached photograph from Receiving Bay 4 at a regional ecommerce distribution center. Prepare a shift-ready safety inspection report for the site EHS manager. Prioritize hazards that need immediate isolation or correction, cite where each condition appears in the image, and avoid reporting anything that is not visibly supported."
                }
            ]
        })


asyncio.run(main())

curl https://api.runware.ai/v1 \
  -H "Authorization: Bearer $RUNWARE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '[
    {
      "taskType": "textInference",
      "taskUUID": "30999b2b-2474-4ed0-87c1-b970299a0a75",
      "model": "openai:gpt@5-nano",
      "settings": {
        "systemPrompt": "You are an experienced warehouse EHS inspection assistant. Analyze only conditions that are directly visible in the supplied image. Do not infer measurements, legal violations, worker intent, or facts outside the frame. Produce a concise professional report with: an executive summary; a numbered findings table containing severity, visible evidence, hazard category, potential consequence, and recommended immediate action; positive safety observations; and items requiring an on-site inspection. Assign severity as Critical, High, Medium, or Low and explain the visual basis. State that the report is a preliminary visual screening and not a substitute for an inspection by a qualified safety professional.",
        "maxTokens": 2500,
        "thinkingLevel": "low"
      },
      "inputs": {
        "images": [
          "https://assets.runware.ai/assets/inputs/510aa7c7-4108-4de0-ba53-5565a164a085.jpg"
        ]
      },
      "messages": [
        {
          "role": "user",
          "content": "Review the attached photograph from Receiving Bay 4 at a regional ecommerce distribution center. Prepare a shift-ready safety inspection report for the site EHS manager. Prioritize hazards that need immediate isolation or correction, cite where each condition appears in the image, and avoid reporting anything that is not visibly supported."
        }
      ]
    }
  ]'

runware run openai:gpt@5-nano \
  settings.systemPrompt="You are an experienced warehouse EHS inspection assistant. Analyze only conditions that are directly visible in the supplied image. Do not infer measurements, legal violations, worker intent, or facts outside the frame. Produce a concise professional report with: an executive summary; a numbered findings table containing severity, visible evidence, hazard category, potential consequence, and recommended immediate action; positive safety observations; and items requiring an on-site inspection. Assign severity as Critical, High, Medium, or Low and explain the visual basis. State that the report is a preliminary visual screening and not a substitute for an inspection by a qualified safety professional." \
  settings.maxTokens=2500 \
  settings.thinkingLevel=low \
  inputs.images.0=https://assets.runware.ai/assets/inputs/510aa7c7-4108-4de0-ba53-5565a164a085.jpg \
  messages.0.role=user \
  messages.0.content="Review the attached photograph from Receiving Bay 4 at a regional ecommerce distribution center. Prepare a shift-ready safety inspection report for the site EHS manager. Prioritize hazards that need immediate isolation or correction, cite where each condition appears in the image, and avoid reporting anything that is not visibly supported."

{
  "taskType": "textInference",
  "taskUUID": "30999b2b-2474-4ed0-87c1-b970299a0a75",
  "model": "openai:gpt@5-nano",
  "settings": {
    "systemPrompt": "You are an experienced warehouse EHS inspection assistant. Analyze only conditions that are directly visible in the supplied image. Do not infer measurements, legal violations, worker intent, or facts outside the frame. Produce a concise professional report with: an executive summary; a numbered findings table containing severity, visible evidence, hazard category, potential consequence, and recommended immediate action; positive safety observations; and items requiring an on-site inspection. Assign severity as Critical, High, Medium, or Low and explain the visual basis. State that the report is a preliminary visual screening and not a substitute for an inspection by a qualified safety professional.",
    "maxTokens": 2500,
    "thinkingLevel": "low"
  },
  "inputs": {
    "images": [
      "https://assets.runware.ai/assets/inputs/510aa7c7-4108-4de0-ba53-5565a164a085.jpg"
    ]
  },
  "messages": [
    {
      "role": "user",
      "content": "Review the attached photograph from Receiving Bay 4 at a regional ecommerce distribution center. Prepare a shift-ready safety inspection report for the site EHS manager. Prioritize hazards that need immediate isolation or correction, cite where each condition appears in the image, and avoid reporting anything that is not visibly supported."
    }
  ]
}

import { createClient } from '@runware/sdk'

const client = await createClient({ apiKey: process.env.RUNWARE_API_KEY })
await client.connect()

const [result] = await client.run({
  model: 'openai:gpt@5-nano',
  settings: {
    systemPrompt: 'You are a meticulous auction-house catalog specialist. Produce commercially usable catalog copy grounded only in the supplied image and explicitly provided facts. Never convert uncertain visual impressions into unsupported claims.',
    maxTokens: 700,
    thinkingLevel: 'low'
  },
  inputs: {
    images: [
      'https://assets.runware.ai/assets/inputs/04b85de8-9034-40ea-abb3-cb2a2a8261ec.jpg'
    ]
  },
  messages: [
    {
      role: 'user',
      content: 'Prepare a publication-ready auction catalog entry for the furniture shown in the attached intake photograph. Use these headings: LOT TITLE, CATALOG DESCRIPTION, VISIBLE CONDITION, and SEARCH KEYWORDS. Write for a reputable regional auction house in a concise, informed, neutral tone. Describe the form, materials, construction clues, hardware, finish, proportions, decorative details, and visible wear. Distinguish direct visual observations from cautious inferences. Do not invent a maker, date, dimensions, provenance, restoration history, wood species, or geographic origin; if uncertain, use qualified language such as “appears to be” or “in the style of.” The lot title should be specific and no longer than 12 words. The catalog description should be 90–130 words, the condition section 50–80 words, and the search keywords should be a single comma-separated line.'
    }
  ]
})

import asyncio
import os

from runware import Runware


async def main():
    async with Runware(api_key=os.environ["RUNWARE_API_KEY"]) as client:
        results = await client.run({
            "model": "openai:gpt@5-nano",
            "settings": {
                "systemPrompt": "You are a meticulous auction-house catalog specialist. Produce commercially usable catalog copy grounded only in the supplied image and explicitly provided facts. Never convert uncertain visual impressions into unsupported claims.",
                "maxTokens": 700,
                "thinkingLevel": "low"
            },
            "inputs": {
                "images": [
                    "https://assets.runware.ai/assets/inputs/04b85de8-9034-40ea-abb3-cb2a2a8261ec.jpg"
                ]
            },
            "messages": [
                {
                    "role": "user",
                    "content": "Prepare a publication-ready auction catalog entry for the furniture shown in the attached intake photograph. Use these headings: LOT TITLE, CATALOG DESCRIPTION, VISIBLE CONDITION, and SEARCH KEYWORDS. Write for a reputable regional auction house in a concise, informed, neutral tone. Describe the form, materials, construction clues, hardware, finish, proportions, decorative details, and visible wear. Distinguish direct visual observations from cautious inferences. Do not invent a maker, date, dimensions, provenance, restoration history, wood species, or geographic origin; if uncertain, use qualified language such as “appears to be” or “in the style of.” The lot title should be specific and no longer than 12 words. The catalog description should be 90–130 words, the condition section 50–80 words, and the search keywords should be a single comma-separated line."
                }
            ]
        })


asyncio.run(main())

curl https://api.runware.ai/v1 \
  -H "Authorization: Bearer $RUNWARE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '[
    {
      "taskType": "textInference",
      "taskUUID": "c6dd806f-90e6-49a7-a489-a44e6f365e19",
      "model": "openai:gpt@5-nano",
      "settings": {
        "systemPrompt": "You are a meticulous auction-house catalog specialist. Produce commercially usable catalog copy grounded only in the supplied image and explicitly provided facts. Never convert uncertain visual impressions into unsupported claims.",
        "maxTokens": 700,
        "thinkingLevel": "low"
      },
      "inputs": {
        "images": [
          "https://assets.runware.ai/assets/inputs/04b85de8-9034-40ea-abb3-cb2a2a8261ec.jpg"
        ]
      },
      "messages": [
        {
          "role": "user",
          "content": "Prepare a publication-ready auction catalog entry for the furniture shown in the attached intake photograph. Use these headings: LOT TITLE, CATALOG DESCRIPTION, VISIBLE CONDITION, and SEARCH KEYWORDS. Write for a reputable regional auction house in a concise, informed, neutral tone. Describe the form, materials, construction clues, hardware, finish, proportions, decorative details, and visible wear. Distinguish direct visual observations from cautious inferences. Do not invent a maker, date, dimensions, provenance, restoration history, wood species, or geographic origin; if uncertain, use qualified language such as “appears to be” or “in the style of.” The lot title should be specific and no longer than 12 words. The catalog description should be 90–130 words, the condition section 50–80 words, and the search keywords should be a single comma-separated line."
        }
      ]
    }
  ]'

runware run openai:gpt@5-nano \
  settings.systemPrompt="You are a meticulous auction-house catalog specialist. Produce commercially usable catalog copy grounded only in the supplied image and explicitly provided facts. Never convert uncertain visual impressions into unsupported claims." \
  settings.maxTokens=700 \
  settings.thinkingLevel=low \
  inputs.images.0=https://assets.runware.ai/assets/inputs/04b85de8-9034-40ea-abb3-cb2a2a8261ec.jpg \
  messages.0.role=user \
  messages.0.content="Prepare a publication-ready auction catalog entry for the furniture shown in the attached intake photograph. Use these headings: LOT TITLE, CATALOG DESCRIPTION, VISIBLE CONDITION, and SEARCH KEYWORDS. Write for a reputable regional auction house in a concise, informed, neutral tone. Describe the form, materials, construction clues, hardware, finish, proportions, decorative details, and visible wear. Distinguish direct visual observations from cautious inferences. Do not invent a maker, date, dimensions, provenance, restoration history, wood species, or geographic origin; if uncertain, use qualified language such as “appears to be” or “in the style of.” The lot title should be specific and no longer than 12 words. The catalog description should be 90–130 words, the condition section 50–80 words, and the search keywords should be a single comma-separated line."

{
  "taskType": "textInference",
  "taskUUID": "c6dd806f-90e6-49a7-a489-a44e6f365e19",
  "model": "openai:gpt@5-nano",
  "settings": {
    "systemPrompt": "You are a meticulous auction-house catalog specialist. Produce commercially usable catalog copy grounded only in the supplied image and explicitly provided facts. Never convert uncertain visual impressions into unsupported claims.",
    "maxTokens": 700,
    "thinkingLevel": "low"
  },
  "inputs": {
    "images": [
      "https://assets.runware.ai/assets/inputs/04b85de8-9034-40ea-abb3-cb2a2a8261ec.jpg"
    ]
  },
  "messages": [
    {
      "role": "user",
      "content": "Prepare a publication-ready auction catalog entry for the furniture shown in the attached intake photograph. Use these headings: LOT TITLE, CATALOG DESCRIPTION, VISIBLE CONDITION, and SEARCH KEYWORDS. Write for a reputable regional auction house in a concise, informed, neutral tone. Describe the form, materials, construction clues, hardware, finish, proportions, decorative details, and visible wear. Distinguish direct visual observations from cautious inferences. Do not invent a maker, date, dimensions, provenance, restoration history, wood species, or geographic origin; if uncertain, use qualified language such as “appears to be” or “in the style of.” The lot title should be specific and no longer than 12 words. The catalog description should be 90–130 words, the condition section 50–80 words, and the search keywords should be a single comma-separated line."
    }
  ]
}

import { createClient } from '@runware/sdk'

const client = await createClient({ apiKey: process.env.RUNWARE_API_KEY })
await client.connect()

const [result] = await client.run({
  model: 'openai:gpt@5-nano',
  settings: {
    systemPrompt: 'You are a senior customer insights analyst. Produce concise, decision-ready business writing grounded only in the supplied evidence. Quantify findings when possible, distinguish facts from hypotheses, and never invent customer statements or metrics.',
    maxTokens: 3500,
    thinkingLevel: 'low'
  },
  messages: [
    {
      role: 'user',
      content: 'Prepare a standalone Q2 2026 churn analysis memo for the leadership team of Northstar Relay, a B2B workflow automation platform.\n\nACCOUNT DATA\n- 42 customers churned, representing $318,000 in annual recurring revenue.\n- Q1 comparison: 29 customers and $201,000 ARR churned.\n- Churned accounts by segment: SMB 27 accounts / $81,000 ARR; Mid-market 12 / $147,000; Enterprise 3 / $90,000.\n- Stated cancellation reasons: missing integrations 12; low adoption 10; budget reduction 8; reliability 6; switched to competitor 4; company closed 2.\n- Median customer age at cancellation: 7 months, down from 11 months in Q1.\n- 26 of 42 churned accounts had fewer than three active users in the final 30 days.\n- 19 had not launched a production workflow within 45 days of purchase.\n- Accounts completing an implementation workshop had 4.8% quarterly logo churn; accounts that did not had 11.9%.\n- Integration-related churn included Salesforce 5, NetSuite 4, Microsoft Dynamics 2, and Workday 1.\n\nSUPPORT AND PRODUCT SIGNALS\n- Failed workflow support tickets rose 34% quarter over quarter.\n- Median first response time improved from 5.2 hours to 3.1 hours.\n- Six reliability-related cancellations followed incidents affecting scheduled workflows between April 18 and May 9.\n- A NetSuite connector beta began June 20; no retention outcome is available yet.\n\nCUSTOMER INTERVIEW EXCERPTS\n1. Operations director, 180-person logistics company, $18,000 ARR: “The demo looked straightforward, but nobody on our team owned implementation. After two months we were still testing sample workflows.”\n2. RevOps lead, 420-person software company, $24,000 ARR: “We needed Salesforce custom-object support. The workaround created more maintenance than it removed.”\n3. COO, 75-person agency, $6,000 ARR: “We cut every tool that was not used weekly. Only two people ever logged in.”\n4. Finance systems manager, 900-person manufacturer, $36,000 ARR: “Two scheduled approval workflows failed during close. Support responded quickly, but confidence was gone.”\n5. IT director, 310-person healthcare services company, $21,000 ARR: “The workshop was valuable, but it happened six weeks after signing. By then the internal champion had moved teams.”\n6. Founder, 40-person ecommerce company, $3,600 ARR: “We moved to a cheaper product. It does less, but it connects to NetSuite without custom work.”\n\nDELIVERABLE\nWrite a board-ready memo in Markdown with these sections:\n1. Executive summary: no more than five bullets.\n2. Q2 churn profile: include a compact table covering account count, ARR lost, quarter-over-quarter change, and segment concentration.\n3. Primary churn drivers: rank the top four drivers and support each with specific metrics and interview evidence.\n4. Interpretation: separate confirmed findings from plausible hypotheses requiring validation.\n5. Recommended actions: provide five prioritized actions, each with an owner, timing, and measurable success indicator. Focus on actions feasible within the next two quarters.\n6. Data gaps: list the most important missing evidence.\n\nUse direct, neutral language. Calculate quarter-over-quarter changes from the supplied figures. Do not claim that workshops, incidents, or integrations caused churn beyond what the evidence supports.'
    }
  ]
})

import asyncio
import os

from runware import Runware


async def main():
    async with Runware(api_key=os.environ["RUNWARE_API_KEY"]) as client:
        results = await client.run({
            "model": "openai:gpt@5-nano",
            "settings": {
                "systemPrompt": "You are a senior customer insights analyst. Produce concise, decision-ready business writing grounded only in the supplied evidence. Quantify findings when possible, distinguish facts from hypotheses, and never invent customer statements or metrics.",
                "maxTokens": 3500,
                "thinkingLevel": "low"
            },
            "messages": [
                {
                    "role": "user",
                    "content": "Prepare a standalone Q2 2026 churn analysis memo for the leadership team of Northstar Relay, a B2B workflow automation platform.\n\nACCOUNT DATA\n- 42 customers churned, representing $318,000 in annual recurring revenue.\n- Q1 comparison: 29 customers and $201,000 ARR churned.\n- Churned accounts by segment: SMB 27 accounts / $81,000 ARR; Mid-market 12 / $147,000; Enterprise 3 / $90,000.\n- Stated cancellation reasons: missing integrations 12; low adoption 10; budget reduction 8; reliability 6; switched to competitor 4; company closed 2.\n- Median customer age at cancellation: 7 months, down from 11 months in Q1.\n- 26 of 42 churned accounts had fewer than three active users in the final 30 days.\n- 19 had not launched a production workflow within 45 days of purchase.\n- Accounts completing an implementation workshop had 4.8% quarterly logo churn; accounts that did not had 11.9%.\n- Integration-related churn included Salesforce 5, NetSuite 4, Microsoft Dynamics 2, and Workday 1.\n\nSUPPORT AND PRODUCT SIGNALS\n- Failed workflow support tickets rose 34% quarter over quarter.\n- Median first response time improved from 5.2 hours to 3.1 hours.\n- Six reliability-related cancellations followed incidents affecting scheduled workflows between April 18 and May 9.\n- A NetSuite connector beta began June 20; no retention outcome is available yet.\n\nCUSTOMER INTERVIEW EXCERPTS\n1. Operations director, 180-person logistics company, $18,000 ARR: “The demo looked straightforward, but nobody on our team owned implementation. After two months we were still testing sample workflows.”\n2. RevOps lead, 420-person software company, $24,000 ARR: “We needed Salesforce custom-object support. The workaround created more maintenance than it removed.”\n3. COO, 75-person agency, $6,000 ARR: “We cut every tool that was not used weekly. Only two people ever logged in.”\n4. Finance systems manager, 900-person manufacturer, $36,000 ARR: “Two scheduled approval workflows failed during close. Support responded quickly, but confidence was gone.”\n5. IT director, 310-person healthcare services company, $21,000 ARR: “The workshop was valuable, but it happened six weeks after signing. By then the internal champion had moved teams.”\n6. Founder, 40-person ecommerce company, $3,600 ARR: “We moved to a cheaper product. It does less, but it connects to NetSuite without custom work.”\n\nDELIVERABLE\nWrite a board-ready memo in Markdown with these sections:\n1. Executive summary: no more than five bullets.\n2. Q2 churn profile: include a compact table covering account count, ARR lost, quarter-over-quarter change, and segment concentration.\n3. Primary churn drivers: rank the top four drivers and support each with specific metrics and interview evidence.\n4. Interpretation: separate confirmed findings from plausible hypotheses requiring validation.\n5. Recommended actions: provide five prioritized actions, each with an owner, timing, and measurable success indicator. Focus on actions feasible within the next two quarters.\n6. Data gaps: list the most important missing evidence.\n\nUse direct, neutral language. Calculate quarter-over-quarter changes from the supplied figures. Do not claim that workshops, incidents, or integrations caused churn beyond what the evidence supports."
                }
            ]
        })


asyncio.run(main())

curl https://api.runware.ai/v1 \
  -H "Authorization: Bearer $RUNWARE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '[
    {
      "taskType": "textInference",
      "taskUUID": "a94c7cc6-ae7b-4351-bc8d-0bd04a2bc825",
      "model": "openai:gpt@5-nano",
      "settings": {
        "systemPrompt": "You are a senior customer insights analyst. Produce concise, decision-ready business writing grounded only in the supplied evidence. Quantify findings when possible, distinguish facts from hypotheses, and never invent customer statements or metrics.",
        "maxTokens": 3500,
        "thinkingLevel": "low"
      },
      "messages": [
        {
          "role": "user",
          "content": "Prepare a standalone Q2 2026 churn analysis memo for the leadership team of Northstar Relay, a B2B workflow automation platform.\n\nACCOUNT DATA\n- 42 customers churned, representing $318,000 in annual recurring revenue.\n- Q1 comparison: 29 customers and $201,000 ARR churned.\n- Churned accounts by segment: SMB 27 accounts / $81,000 ARR; Mid-market 12 / $147,000; Enterprise 3 / $90,000.\n- Stated cancellation reasons: missing integrations 12; low adoption 10; budget reduction 8; reliability 6; switched to competitor 4; company closed 2.\n- Median customer age at cancellation: 7 months, down from 11 months in Q1.\n- 26 of 42 churned accounts had fewer than three active users in the final 30 days.\n- 19 had not launched a production workflow within 45 days of purchase.\n- Accounts completing an implementation workshop had 4.8% quarterly logo churn; accounts that did not had 11.9%.\n- Integration-related churn included Salesforce 5, NetSuite 4, Microsoft Dynamics 2, and Workday 1.\n\nSUPPORT AND PRODUCT SIGNALS\n- Failed workflow support tickets rose 34% quarter over quarter.\n- Median first response time improved from 5.2 hours to 3.1 hours.\n- Six reliability-related cancellations followed incidents affecting scheduled workflows between April 18 and May 9.\n- A NetSuite connector beta began June 20; no retention outcome is available yet.\n\nCUSTOMER INTERVIEW EXCERPTS\n1. Operations director, 180-person logistics company, $18,000 ARR: “The demo looked straightforward, but nobody on our team owned implementation. After two months we were still testing sample workflows.”\n2. RevOps lead, 420-person software company, $24,000 ARR: “We needed Salesforce custom-object support. The workaround created more maintenance than it removed.”\n3. COO, 75-person agency, $6,000 ARR: “We cut every tool that was not used weekly. Only two people ever logged in.”\n4. Finance systems manager, 900-person manufacturer, $36,000 ARR: “Two scheduled approval workflows failed during close. Support responded quickly, but confidence was gone.”\n5. IT director, 310-person healthcare services company, $21,000 ARR: “The workshop was valuable, but it happened six weeks after signing. By then the internal champion had moved teams.”\n6. Founder, 40-person ecommerce company, $3,600 ARR: “We moved to a cheaper product. It does less, but it connects to NetSuite without custom work.”\n\nDELIVERABLE\nWrite a board-ready memo in Markdown with these sections:\n1. Executive summary: no more than five bullets.\n2. Q2 churn profile: include a compact table covering account count, ARR lost, quarter-over-quarter change, and segment concentration.\n3. Primary churn drivers: rank the top four drivers and support each with specific metrics and interview evidence.\n4. Interpretation: separate confirmed findings from plausible hypotheses requiring validation.\n5. Recommended actions: provide five prioritized actions, each with an owner, timing, and measurable success indicator. Focus on actions feasible within the next two quarters.\n6. Data gaps: list the most important missing evidence.\n\nUse direct, neutral language. Calculate quarter-over-quarter changes from the supplied figures. Do not claim that workshops, incidents, or integrations caused churn beyond what the evidence supports."
        }
      ]
    }
  ]'

runware run openai:gpt@5-nano \
  settings.systemPrompt="You are a senior customer insights analyst. Produce concise, decision-ready business writing grounded only in the supplied evidence. Quantify findings when possible, distinguish facts from hypotheses, and never invent customer statements or metrics." \
  settings.maxTokens=3500 \
  settings.thinkingLevel=low \
  messages.0.role=user \
  messages.0.content="Prepare a standalone Q2 2026 churn analysis memo for the leadership team of Northstar Relay, a B2B workflow automation platform.

ACCOUNT DATA
- 42 customers churned, representing \$318,000 in annual recurring revenue.
- Q1 comparison: 29 customers and \$201,000 ARR churned.
- Churned accounts by segment: SMB 27 accounts / \$81,000 ARR; Mid-market 12 / \$147,000; Enterprise 3 / \$90,000.
- Stated cancellation reasons: missing integrations 12; low adoption 10; budget reduction 8; reliability 6; switched to competitor 4; company closed 2.
- Median customer age at cancellation: 7 months, down from 11 months in Q1.
- 26 of 42 churned accounts had fewer than three active users in the final 30 days.
- 19 had not launched a production workflow within 45 days of purchase.
- Accounts completing an implementation workshop had 4.8% quarterly logo churn; accounts that did not had 11.9%.
- Integration-related churn included Salesforce 5, NetSuite 4, Microsoft Dynamics 2, and Workday 1.

SUPPORT AND PRODUCT SIGNALS
- Failed workflow support tickets rose 34% quarter over quarter.
- Median first response time improved from 5.2 hours to 3.1 hours.
- Six reliability-related cancellations followed incidents affecting scheduled workflows between April 18 and May 9.
- A NetSuite connector beta began June 20; no retention outcome is available yet.

CUSTOMER INTERVIEW EXCERPTS
1. Operations director, 180-person logistics company, \$18,000 ARR: “The demo looked straightforward, but nobody on our team owned implementation. After two months we were still testing sample workflows.”
2. RevOps lead, 420-person software company, \$24,000 ARR: “We needed Salesforce custom-object support. The workaround created more maintenance than it removed.”
3. COO, 75-person agency, \$6,000 ARR: “We cut every tool that was not used weekly. Only two people ever logged in.”
4. Finance systems manager, 900-person manufacturer, \$36,000 ARR: “Two scheduled approval workflows failed during close. Support responded quickly, but confidence was gone.”
5. IT director, 310-person healthcare services company, \$21,000 ARR: “The workshop was valuable, but it happened six weeks after signing. By then the internal champion had moved teams.”
6. Founder, 40-person ecommerce company, \$3,600 ARR: “We moved to a cheaper product. It does less, but it connects to NetSuite without custom work.”

DELIVERABLE
Write a board-ready memo in Markdown with these sections:
1. Executive summary: no more than five bullets.
2. Q2 churn profile: include a compact table covering account count, ARR lost, quarter-over-quarter change, and segment concentration.
3. Primary churn drivers: rank the top four drivers and support each with specific metrics and interview evidence.
4. Interpretation: separate confirmed findings from plausible hypotheses requiring validation.
5. Recommended actions: provide five prioritized actions, each with an owner, timing, and measurable success indicator. Focus on actions feasible within the next two quarters.
6. Data gaps: list the most important missing evidence.

Use direct, neutral language. Calculate quarter-over-quarter changes from the supplied figures. Do not claim that workshops, incidents, or integrations caused churn beyond what the evidence supports."

{
  "taskType": "textInference",
  "taskUUID": "a94c7cc6-ae7b-4351-bc8d-0bd04a2bc825",
  "model": "openai:gpt@5-nano",
  "settings": {
    "systemPrompt": "You are a senior customer insights analyst. Produce concise, decision-ready business writing grounded only in the supplied evidence. Quantify findings when possible, distinguish facts from hypotheses, and never invent customer statements or metrics.",
    "maxTokens": 3500,
    "thinkingLevel": "low"
  },
  "messages": [
    {
      "role": "user",
      "content": "Prepare a standalone Q2 2026 churn analysis memo for the leadership team of Northstar Relay, a B2B workflow automation platform.\n\nACCOUNT DATA\n- 42 customers churned, representing $318,000 in annual recurring revenue.\n- Q1 comparison: 29 customers and $201,000 ARR churned.\n- Churned accounts by segment: SMB 27 accounts / $81,000 ARR; Mid-market 12 / $147,000; Enterprise 3 / $90,000.\n- Stated cancellation reasons: missing integrations 12; low adoption 10; budget reduction 8; reliability 6; switched to competitor 4; company closed 2.\n- Median customer age at cancellation: 7 months, down from 11 months in Q1.\n- 26 of 42 churned accounts had fewer than three active users in the final 30 days.\n- 19 had not launched a production workflow within 45 days of purchase.\n- Accounts completing an implementation workshop had 4.8% quarterly logo churn; accounts that did not had 11.9%.\n- Integration-related churn included Salesforce 5, NetSuite 4, Microsoft Dynamics 2, and Workday 1.\n\nSUPPORT AND PRODUCT SIGNALS\n- Failed workflow support tickets rose 34% quarter over quarter.\n- Median first response time improved from 5.2 hours to 3.1 hours.\n- Six reliability-related cancellations followed incidents affecting scheduled workflows between April 18 and May 9.\n- A NetSuite connector beta began June 20; no retention outcome is available yet.\n\nCUSTOMER INTERVIEW EXCERPTS\n1. Operations director, 180-person logistics company, $18,000 ARR: “The demo looked straightforward, but nobody on our team owned implementation. After two months we were still testing sample workflows.”\n2. RevOps lead, 420-person software company, $24,000 ARR: “We needed Salesforce custom-object support. The workaround created more maintenance than it removed.”\n3. COO, 75-person agency, $6,000 ARR: “We cut every tool that was not used weekly. Only two people ever logged in.”\n4. Finance systems manager, 900-person manufacturer, $36,000 ARR: “Two scheduled approval workflows failed during close. Support responded quickly, but confidence was gone.”\n5. IT director, 310-person healthcare services company, $21,000 ARR: “The workshop was valuable, but it happened six weeks after signing. By then the internal champion had moved teams.”\n6. Founder, 40-person ecommerce company, $3,600 ARR: “We moved to a cheaper product. It does less, but it connects to NetSuite without custom work.”\n\nDELIVERABLE\nWrite a board-ready memo in Markdown with these sections:\n1. Executive summary: no more than five bullets.\n2. Q2 churn profile: include a compact table covering account count, ARR lost, quarter-over-quarter change, and segment concentration.\n3. Primary churn drivers: rank the top four drivers and support each with specific metrics and interview evidence.\n4. Interpretation: separate confirmed findings from plausible hypotheses requiring validation.\n5. Recommended actions: provide five prioritized actions, each with an owner, timing, and measurable success indicator. Focus on actions feasible within the next two quarters.\n6. Data gaps: list the most important missing evidence.\n\nUse direct, neutral language. Calculate quarter-over-quarter changes from the supplied figures. Do not claim that workshops, incidents, or integrations caused churn beyond what the evidence supports."
    }
  ]
}