How Stoka works

An artifact is the unit Stoka organizes around: a permalinked, author-signed, shareable piece of AI work. It has a kind, and the kind tells the platform how to render it, retrieve it, and relate it to the rest of your library.

• conversation A chat, pasted in or captured from a connected tool.
• spec A design doc, plan, or written rationale.
• essay Long-form writing with a voice behind it.
• prompt A reusable instruction with variables.
• note Anything short-form worth keeping.
• codebase A snapshot of code, shareable anonymously.
• link An outbound pointer with context attached.

Every kind lives in the same library, indexed by the same retrieval layer, governed by the same sharing primitives. One primitive, many shapes.

A good artifact is not just “the raw transcript, but searchable.” Stoka renders work in layers so the reader can stop at the useful depth instead of being forced straight into the entire chat log.

1. Recipe

   The scan layer: what this is, why it worked, and what to try next.

2. Method

   What was attempted, what failed, and where the thing finally clicked.

3. Source

   The specific exchanges Stoka pulled from, preserved in order and still attributable.

4. Raw

   The full transcript, only when the author explicitly allows that layer to travel.

Phase 1 starts with paste and file upload because that is the shortest route to a solo tool that is genuinely useful. Drop in a conversation, a spec, or a raw export. Stoka detects format, structures the work, flags what should be scrubbed, and turns it into something you can actually keep.

Retrieval is the actual test. If “that thing about caching embeddings” does not come back in seconds, capture was just filing. Stoka indexes your library so you can search it in natural language, traverse laterally from one artifact to adjacent ones, and recover your own past thinking without remembering exact filenames or dates.

Default visibility is private. The main action is send-to-handle, not publish-to-public. A share grants access. A thread exists to carry context around the artifact, not to replace it. Pseudonyms are the identity primitive, which means the platform never asks you to collapse into one public self just to let work move.

Stoka ships in phases, and each phase must stand on its own before the next opens.

1. Phase 1

   Personal library

   Capture, retrieve, pseudonyms. Must feel amazing alone before anyone is invited.

2. Phase 2

   Share-to-handle and DMs

   Rich artifact embeds, inbox, opt-in notifications. Growth through trusted hops, not virality loops.

3. Phase 3

   Opt-in public layer

   A discovery tier you choose into per pseudonym. The directory is the exception, not the default surface.

Stoka-bot is the site's editorial nervous system. Same identity, same constraints, same voice across artifact extraction, personal-library retrieval, public discovery, and moderation later on. Different context, same brain. That is how the product stays coherent instead of feeling like six disconnected features.

This matters because the platform is editorial by design. It does not just store work. It frames, structures, and surfaces it.

Debono, Destilo, Distilio, StokaTerminal, Screwit, and GhostRite are the six live reference integrations. They are not the category. They are the proof surfaces where the category gets tested against real user workflows.

See the integrations →

The full platform mission, phase gates, the artifact model, and the messaging layer are written up in the specs directory. Start with the README if you want the deep version.