Imprint

Website: imprint.niclydon.dev

Public-first identity and expression profile compiler.

Imprint analyzes human-authored artifacts such as writing, messages, transcripts, notes, and documents, then compiles structured expression profiles that downstream systems can use without reading the raw private corpus.

🚧 Pre-release / active development

Imprint is open-source and public-first by design, but not yet ready for production use.

Current focus:

1. Packaging and onboarding for a public developer preview
2. Public landing site at imprint.niclydon.dev
3. Keeping all examples synthetic and all private corpora out of git

This path uses only the committed synthetic corpus. It does not require private services, credentials, model providers, or network connectors.

git clone https://github.com/niclydon/imprint.git
cd imprint
python -m venv .venv
source .venv/bin/activate
pip install -e ".[test]"

imprint --help
imprint connectors-dry-run --config imprint.config.example.yaml
imprint example

sed -n '1,80p' exports/synthetic-demo/what-imprint-learned.md
sed -n '1,80p' exports/synthetic-demo/profile.md

Expected example outputs are written under exports/synthetic-demo/, which is ignored by git. The generated files include canonical JSON, Markdown, first-run summary, Mosvera expression overlay, and human CLI consumer-contract examples.

For a fuller walkthrough, see docs/QUICKSTART.md.

Imprint turns evidence into reusable profiles:

configured sources
  -> harvest
  -> classify
  -> extract signals
  -> compile profile
  -> export contracts

It is designed to produce evidence-backed expression profiles that can support:

• writing tools,
• publishing systems,
• agent personas,
• aesthetic/identity packs,
• profile comparison,
• drift analysis,
• and first-run “What Imprint Learned” reports.

Imprint is not:

• a memory system,
• a data lake,
• a publishing platform,
• a model router,
• a voice-cloning service,
• a digital twin,
• a personality test,
• or a diagnostic system.

Imprint observes expression patterns. It does not diagnose people or infer mental health states.

• Profile + Artifact: a profile describes patterns; an artifact uses those patterns.
• Evidence first: every claim should answer “why do we believe this?”
• Confidence is not truth: confidence describes support strength, not certainty about a person.
• Store less than you can: durable signals matter more than maximal capture.
• Local-first: private corpora should remain under user control by default.
• Public-safe: examples, tests, and docs use synthetic data only.

• docs/README.md — documentation map
• docs/QUICKSTART.md — copy-paste synthetic onboarding path
• docs/INSTALL.md — clean install and local verification
• docs/PROJECT_STRATEGY.md — project strategy
• docs/PRODUCT_POSITIONING.md — what Imprint is and is not
• docs/PROFILE_THEORY.md — expression profile theory and claim boundaries
• docs/SCHEMA.md — canonical schema overview
• docs/EXPORT_FORMATS.md — public-safe export contracts
• docs/MODEL_PROVIDER_POLICY.md — BYOM/BYOP model policy
• docs/PRIVACY_AND_LOCAL_MODE.md — privacy and local-first posture
• docs/RELEASE_CHECKLIST.md — v0.1.0 release checklist

This repository is designed to be public from the beginning. Real corpora, secrets, local configs, generated private profiles, and private exports must stay out of git.

Use:

• .env.example
• imprint.config.example.yaml
• synthetic examples only

Never commit:

• .env
• imprint.config.yaml
• real messages
• real emails
• real transcripts
• real profile exports
• private database dumps

Raw local research and private notes belong under private/, which is ignored by git.

MIT. See LICENSE.