# LeMUR — Run LLMs Over AssemblyAI Transcripts

> LeMUR runs Claude / GPT prompts over AssemblyAI transcripts already in context. Summaries, Q&A, action items, custom JSON extraction.

## Install

Save the content below to `.claude/skills/` or append to your `CLAUDE.md`:

## Quick Use

1. Transcribe with `aai.Transcriber().transcribe(...)`
2. Call `transcript.lemur.summarize / question / task / action_items`
3. Pick `final_model` per cost/quality tradeoff

---

## Intro

LeMUR (Leveraging Large Language Models to Understand Recognized Speech) is AssemblyAI's transcript-LLM bridge — once a transcript exists in your account, you can run Claude or GPT prompts against it without re-uploading or chunking. Endpoints: summary, Q&A, action items, custom prompt. Best for: meeting recap automation, call center QA, podcast show notes, any post-transcription analysis. Works with: assemblyai Python/Node SDK + LeMUR HTTP endpoints. Setup time: 5 minutes after a transcript exists.

---

### Summary endpoint

```python
import assemblyai as aai
aai.settings.api_key = ASSEMBLYAI_KEY

transcript = aai.Transcriber().transcribe("call.mp3")

summary = transcript.lemur.summarize(
    final_model=aai.LemurModel.claude3_5_sonnet,
    context="This is a customer support call about a missed refund.",
    answer_format="3 bullet points",
)
print(summary.response)
```

### Custom prompt (most flexible)

```python
prompt = '''
You are a call center QA analyst. Score this support call on:
- Empathy (0-10)
- Resolution clarity (0-10)
- Compliance: was the agent's name stated, was a case number provided?

Return strict JSON with these fields plus a 'notes' string under 200 words.
'''

result = transcript.lemur.task(
    prompt=prompt,
    final_model=aai.LemurModel.claude3_5_sonnet,
    temperature=0.0,
    max_output_size=600,
)
import json
print(json.loads(result.response))
```

### Q&A endpoint (multi-question)

```python
qa = transcript.lemur.question(
    questions=[
        aai.LemurQuestion(question="What was the customer's main complaint?"),
        aai.LemurQuestion(question="Did the agent offer a refund? If yes, how much?"),
        aai.LemurQuestion(question="What's the recommended next action?", answer_format="one sentence"),
    ],
    final_model=aai.LemurModel.claude3_5_sonnet,
)
for r in qa.response:
    print(r.question, "→", r.answer)
```

### Action items

```python
action_items = transcript.lemur.action_items(
    final_model=aai.LemurModel.claude3_5_sonnet,
    context="Internal product planning meeting.",
)
print(action_items.response)
```

### Available models

| Model | Best for |
|---|---|
| `claude3_5_sonnet` | Default — best quality, balanced cost |
| `claude3_haiku` | Cheap, fast for short summaries |
| `claude3_opus` | Top quality, slowest, highest cost |
| `default` | AssemblyAI-tuned fast model |

---

### FAQ

**Q: Why use LeMUR instead of feeding transcript to Claude myself?**
A: Three reasons: (1) the transcript stays in AssemblyAI's secure data plane — no re-upload of potentially-PII content; (2) you skip the chunking + context management plumbing; (3) it's one billing invoice. For one-off scripts, calling Claude directly is fine; for production analyze-every-call flows, LeMUR is simpler.

**Q: Can I run LeMUR on multiple transcripts at once?**
A: Yes — `aai.Lemur().task(transcript_ids=[id1, id2, id3], prompt=...)`. Useful for weekly call-portfolio analysis. 100 transcripts max per call.

**Q: Does LeMUR support tool calls?**
A: Not yet — LeMUR is text-in/text-out. For tool use, fetch the transcript, then pass it to your own Claude/OpenAI call with tools enabled.

---

## Source & Thanks

> Built by [AssemblyAI](https://github.com/AssemblyAI). LeMUR docs at [assemblyai.com/docs/lemur](https://assemblyai.com/docs/lemur).
>
> [AssemblyAI/assemblyai-python-sdk](https://github.com/AssemblyAI/assemblyai-python-sdk)

---

<!-- ZH -->

## 快速使用

1. 用 `aai.Transcriber().transcribe(...)` 转录
2. 调 `transcript.lemur.summarize / question / task / action_items`
3. 按成本/质量折中选 `final_model`

---

## 简介

LeMUR（Leveraging Large Language Models to Understand Recognized Speech）是 AssemblyAI 的转录-LLM 桥 —— 转录在你账户里之后，可以对它跑 Claude 或 GPT prompt，不用重新上传或分块。Endpoint：summary、Q&A、action items、custom prompt。适合会议复盘自动化、呼叫中心 QA、播客 show notes、任何转录后分析。兼容 assemblyai Python/Node SDK + LeMUR HTTP endpoint。装机时间：转录存在后 5 分钟。

---

### Summary endpoint

```python
import assemblyai as aai
aai.settings.api_key = ASSEMBLYAI_KEY

transcript = aai.Transcriber().transcribe("call.mp3")

summary = transcript.lemur.summarize(
    final_model=aai.LemurModel.claude3_5_sonnet,
    context="这是关于一笔未到账退款的客服通话。",
    answer_format="3 个要点",
)
print(summary.response)
```

### 自定义 prompt（最灵活）

```python
prompt = '''
你是呼叫中心 QA 分析师。给这次客服通话打分：
- 同理心（0-10）
- 解决清晰度（0-10）
- 合规：agent 报过名字吗？有给案件编号吗？

返回严格 JSON 含这些字段加 'notes' 字符串 <200 字。
'''

result = transcript.lemur.task(
    prompt=prompt,
    final_model=aai.LemurModel.claude3_5_sonnet,
    temperature=0.0,
    max_output_size=600,
)
import json
print(json.loads(result.response))
```

### Q&A endpoint（多问题）

```python
qa = transcript.lemur.question(
    questions=[
        aai.LemurQuestion(question="客户主要抱怨什么？"),
        aai.LemurQuestion(question="客服提了退款吗？多少？"),
        aai.LemurQuestion(question="下一步建议？", answer_format="一句话"),
    ],
    final_model=aai.LemurModel.claude3_5_sonnet,
)
for r in qa.response:
    print(r.question, "→", r.answer)
```

### Action items

```python
action_items = transcript.lemur.action_items(
    final_model=aai.LemurModel.claude3_5_sonnet,
    context="内部产品规划会。",
)
print(action_items.response)
```

### 可用模型

| 模型 | 最佳用途 |
|---|---|
| `claude3_5_sonnet` | 默认 —— 质量最好、成本均衡 |
| `claude3_haiku` | 短摘要便宜快 |
| `claude3_opus` | 顶级质量，最慢，最贵 |
| `default` | AssemblyAI 调过的快速模型 |

---

### FAQ

**Q: 为啥用 LeMUR 不自己把转录喂 Claude？**
A: 三个原因：(1) 转录留在 AssemblyAI 安全数据面 —— 不用重新上传可能含 PII 的内容；(2) 跳过分块 + 上下文管理水管；(3) 一张账单。一次性脚本直接调 Claude 没问题；生产分析每通电话流程 LeMUR 更简单。

**Q: 能一次性对多个转录跑 LeMUR 吗？**
A: 能 —— `aai.Lemur().task(transcript_ids=[id1, id2, id3], prompt=...)`。适合每周通话组合分析。每次最多 100 个转录。

**Q: LeMUR 支持 tool call 吗？**
A: 还不支持 —— LeMUR 是文本进文本出。要 tool use 就先拉转录再传给你自己开 tools 的 Claude/OpenAI 调用。

---

## 来源与感谢

> Built by [AssemblyAI](https://github.com/AssemblyAI). LeMUR docs at [assemblyai.com/docs/lemur](https://assemblyai.com/docs/lemur).
>
> [AssemblyAI/assemblyai-python-sdk](https://github.com/AssemblyAI/assemblyai-python-sdk)


---
Source: https://tokrepo.com/en/workflows/lemur-run-llms-over-assemblyai-transcripts
Author: AssemblyAI