Developer ToolsFree generatorPDF, Excel, Word export

Ollama Modelfile Generator for an ERP Assistant

Ollama lets a team stand up a local model in minutes, which makes it a good pilot tool for ERP questions. A Modelfile captures the base model, parameters and house rules so everyone gets the same behavior.

Fill in

Generated code
Download:

Modelfile

FROM qwen3:14b

PARAMETER num_ctx 8192
PARAMETER temperature 0.1
PARAMETER top_p 0.9
PARAMETER repeat_penalty 1.05

SYSTEM """
You are the internal assistant for Acme Precision Ltd, answering questions about data from Infor SyteLine.
You only use the context supplied in the conversation. You are read-only: never offer to change, create or delete ERP records.
If asked to perform a write action, explain that a person must do it in Infor SyteLine.
Quote the table or report name you used. Never guess a part number or a price. Say 'not found in the data provided' when the context lacks the answer.
"""

Create and run

ollama create erp-assistant -f Modelfile
ollama run erp-assistant

Call it from a script or integration

curl http://localhost:11434/api/chat -d '{
  "model": "erp-assistant",
  "stream": false,
  "messages": [
    {"role": "user", "content": "Summarise open purchase orders past due."}
  ]
}'

Verify parameter names against your Ollama version. A bigger num_ctx uses more memory; raise it only as far as your prompts need. A system prompt is guidance, not a security control: enforce read-only access in the database or API role, not in the prompt.

Pilot to production plan

An optional outline for moving a local Ollama pilot to a governed deployment.

Optional. The tool and every export work without this. No spam.

How to use it

Choose a base model you already pulled with ollama pull, give the assistant a name, and describe your ERP and company. Paste the generated Modelfile into a file called Modelfile, run the create command and then chat. The third block shows the REST call that integrations use, so your scripts can query the same model on port 11434.

Why these parameter choices

A low temperature keeps answers about numbers and statuses consistent from run to run. The context window must be large enough to hold the retrieved ERP rows or document excerpts you paste in, but Ollama defaults can be small, so setting num_ctx explicitly avoids silent truncation. A mild repeat penalty reduces looping on tabular output.

  • temperature near 0.1 for factual lookups
  • num_ctx sized to your retrieved context
  • SYSTEM text sets scope and refusal behavior
  • One Modelfile per use case keeps prompts clean

Common mistakes

The biggest error is treating the system prompt as a security boundary. A model can be talked out of its instructions, so permissions belong in the database role or API scope. Another is leaving the default context size and wondering why long reports are cut off. Finally, pilots that stay on a laptop model should be sized properly before many users depend on them.

Frequently Asked Questions

Does the Modelfile give the model access to my ERP?

No. It only configures the model. Access to ERP data comes from whatever retrieval or tool layer you connect, which should use a read-only account with the same permissions as the asking user.

What context size should I set?

Set it to cover your system prompt, the retrieved rows or passages and the answer. Many pilots use 8192 to 16384. Larger windows consume more memory and slow responses.

Can I run Ollama without internet?

Yes after the model files are present. Pull models on a connected machine and move them across using your approved process, or copy the Ollama models directory to the isolated host.

When should I move from Ollama to vLLM?

When several users hit the endpoint at once or you need higher throughput on server GPUs. Ollama is excellent for pilots and single-user tools.

Netray can take a working Ollama pilot to a governed, read-only ERP assistant on your own servers, including retrieval and audit logging.

Talk to an ExpertExplore More Free Tools

Related Tools

Developer Tools

vLLM Docker Compose Generator for a Private ERP LLM

Fill in the model, context length and GPU count to get a docker-compose file that serves an OpenAI-compatible endpoint inside your network.

On-Prem AI

Open-Weight LLM License Terms Reference

Look up which license each popular open-weight model family ships under and what to check before running it on-prem against ERP data.

AI Agents & Automation

Text-to-SQL Guardrails Checklist for ERP Data

Check that an LLM that writes SQL over your ERP is limited by database permissions, query validation, cost limits and audit trails.

Compliance & Contracts

AI Data Classification Handling Matrix Worksheet

Decide per data class whether it may be sent to a hosted LLM, an on-prem model or no model, and who approves exceptions.

More Developer Tools tools

See all
Back to All Tools