Local LLM (small models) prompts
Short, strict prompts tuned for small 1B-9B local models running in Ollama or LM Studio.
Pick a model for your hardware with the GPU model picker on Lila Forge Local AI.
Small-model friendly summariser
Tuned for 3B-9B local models (Gemma 4 E4B, Qwen3.5 9B, Llama 3.2 3B)
Variables: text
Summarise the text below.
Rules:
- Exactly 5 bullet points.
- Each bullet is one sentence, under 20 words.
- Only use facts from the text.
Text:
"""
{{text}}
"""
Summary: Structured JSON extraction (small models)
Use Ollama's format/JSON-schema option for reliability
Variables: text
Extract the fields below from the text. Reply with JSON only, no explanation.
Fields:
- name (string)
- email (string or null)
- company (string or null)
- request (string, max 15 words)
Example output:
{"name": "Sam Lee", "email": "sam@example.com", "company": null, "request": "wants a quote for 20 licences"}
Text:
"""
{{text}}
""" Classify into fixed labels
Reliable on 1B-4B models with temperature 0
Variables: labelsmessage
Classify the message into exactly one label.
Labels: {{labels}}
Reply with the label only.
Message: {{message}}
Label: Local RAG answer prompt
For chat models in a local RAG pipeline
Variables: retrieved_chunksquestion
Use the context to answer the question. If the answer is not in the context, say "I don't know based on these documents."
Quote the sentence you used in "quotes".
Context:
{{retrieved_chunks}}
Question: {{question}}
Answer: Code explainer for small coder models
Qwen2.5-Coder 7B, Devstral, Qwen3-Coder
Variables: languagecode
Explain what this {{language}} function does.
```{{language}}
{{code}}
```
Answer in this format:
Purpose: (one sentence)
Inputs: (list)
Output: (one sentence)
Steps: (numbered, max 6)
Possible bugs: (list, or "none spotted") Commit message from diff (tiny model)
Works on 1.5B-3B coder models; pipe `git diff --staged` in
Variables: diff
Write a git commit message for this diff.
Format: one line, imperative mood, max 60 characters. Then a blank line and up to 3 bullet points.
Do not mention files that are not in the diff.
{{diff}} Offline email triage
7B+ local model; keeps your inbox private
Variables: emails
For each email below, output one line: <number> | <urgent/normal/low> | <reply needed: yes/no> | <one-line summary>.
Urgent = needs action today, from a real person, about money, deadlines or problems.
Emails:
{{emails}} Few-shot style mimic (small models)
Give 2-3 examples; small models copy patterns well
Variables: output_typeexample_1example_2topic
Write a {{output_type}} in the same style as the examples.
Example 1:
{{example_1}}
Example 2:
{{example_2}}
Now write one about: {{topic}} Get all 209 prompts + agent workflows
Every prompt as Markdown and JSON, 10 multi-step agent workflows, and 12 ready-made slash commands for Cursor and Claude Code. Instant zip download, one-time purchase.
New prompts every week
One short email with the best new prompts and agent workflows we've tested. Free; unsubscribe any time.