path to freedom ◆ local AI ◆ open-weight literacy

Local and Open AI

A practical starting point for people who want to understand local models, protect their data, read model licenses, and help build community-owned AI learning resources.

start here

This is education first, training later.

DigitalQuill is exploring a community model path, but the first step is not claiming we have trained a model. The first step is helping people run small local models, judge them through the Trinity lens, and contribute safe examples that could support future evals or fine-tuning.

Read the Trinity guide View community model files
trinity model scorecard

Every model or tool should pass the Trinity filter.

Outcome

Does this help a real person build, learn, decide, or ship something useful?

  • Can a beginner try it?
  • Does it solve a real workflow problem?
  • Can we test whether it helped?

Agency

Does this increase ownership, portability, privacy, and independence?

  • Can the user export or leave?
  • Does the work stay understandable?
  • Does the user keep control?

AI Hygiene

Are license, safety, data handling, evals, and limitations clear?

  • Is the license reviewed?
  • Are private submissions blocked?
  • Are results evaluated honestly?
community path

How we get there without pretending.

Step 1Learn local inference. Run small open-weight models locally and understand what they are good and bad at.
Step 2Build evals first. Create test cases for Trinity behavior before training anything.
Step 3Collect safe examples. Use structured, fictionalized, consented examples instead of raw private chats.
Step 4Review and redact. No example enters a dataset until it passes privacy, copyright, safety, and quality checks.
Step 5Fine-tune small. Try a narrow LoRA/QLoRA adapter only after the dataset and evals are clean.
Step 6Publish honestly. Share what improved, what failed, what hardware was needed, and what remains unsafe or unproven.
privacy and security

Freedom is not real if the data is reckless.

Do not submit private chats

Community examples should be structured and fictionalized. Raw chats can contain private names, account details, customer information, or secrets.

Do not expose local servers

Local AI tools should stay local unless the user intentionally secures and publishes them. A local model endpoint exposed to the internet can become a serious security problem.

Review licenses

Many models are open-weight but not fully open-source. Read the license before using a model commercially or publishing derivatives.

What we are not claiming yet

We are not claiming that DigitalQuill has trained its own model. We are building the safe foundation first: education, contribution rules, evals, seed examples, and a responsible path toward a small future fine-tune.

contribute safely

What the community can contribute now.

Prompt examples

Short prompts that teach project memory, scope control, verification, or Trinity reasoning.

Eval cases

Small tests that show whether a model can protect Outcome, Agency, and AI Hygiene.

Model notes

Hardware, speed, quality, privacy, and license notes from trying local/open-weight models.

Contribution rules Example template