Privacy that is structural, not promised.
Most tools promise not to look at your data. Indelis is built so there is nothing to look at: the stack runs on your hardware, the models can be local, and the defaults are the most private option available.
Runs on your machine
Docker quickstart on your hardware. No telemetry, no phone-home. Uptime and backups are yours to control — and yours to own.
Local models first
A resident local model set covers the platform; heavier models load on demand. Cloud is an explicit per-user choice with your own key — never a silent default.
Most-private defaults
A fresh instance seeds consent at the most private setting. Widening — enabling cloud models, sharing — is an explicit act, per user, reversible.
Nothing trains on you
Your notes never become training data. There is no analytics pipeline on content, no engagement profiling — the product has no feed to optimise.
Leave whole
The vault is Markdown with frontmatter. Full export at any moment; the composition — notes, bands, connections — stays yours.
Small attack surface
Static-first marketing site, self-host product with per-tenant isolation, CSP and rate limits. No comment database, no third-party embeds.
Where the models run
Every LLM-touched path respects the same rule: locality is a user-level choice the router must honour.
Ollama-compatible models on your box. The platform is engineered to be fully usable this way.
Bring your own API key for cloud models. Keys are per-user, never shared between users of a tenant.
Pin classes of work to local; a cloud pin never overrides a local-preferred class on a mixed pool.
If the privacy of the tool depends on trusting the vendor, it is not privacy — it is a subscription to a promise. Self-host removes the vendor from the equation.