explAIn Documentation

Product Vision

4 min read

The thesis

Most “AI in chat” products do one of two things: they write your messages for you and pretend you wrote them, or they bolt a chatbot onto a sidebar. Both treat the LLM as an authorial replacement.

explAIn takes the opposite position. The LLM is scaffolding for human conversation, not a substitute for it. Two humans are still talking to each other. The LLMs are visible helpers — and because they are visible, they can be more useful than a hidden ghostwriter ever could be.

Every conversation has three roles:

  1. You — the human author.
  2. Your advisor — a personal LLM you configure. It reads your draft, proposes a revision, and you decide what to send. It is private to you; your peer never sees your advisor’s output.
  3. The mediator — a shared LLM with a persona both parties have agreed to. It processes your message after you send it, and its output is visible to both parties.

The transparency contract

The single non-negotiable rule of the product:

Every transformation an LLM makes to a message is visible to the parties affected by it.

Concretely:

  • The advisor’s suggestion is visible to its owner. The peer never sees the advisor’s suggestion — the advisor is private counsel.
  • The mediator’s transformation, annotation, or block is visible to both parties, including the mediator’s stated reasoning.
  • Any persona switch is logged as a system event in the conversation, with a timestamp and who proposed it.
  • Token usage, model, and provider for each LLM call are recorded (internal in v0.1; exposed later for the paid quota tier).

The “show your work” UI is the literal embodiment of this contract. Every message carries a collapsible panel showing the original draft, the advisor suggestion, the sent text, and the mediator output side-by-side.

What counts as a regression

The contract is so central that it defines what a regression is:

  • Hiding the mediator’s transformation, action, or reasoning from either party breaks the core value proposition.
  • Hiding the advisor’s suggestion from its owner breaks it too.
  • Showing the advisor’s suggestion to the peer also breaks it — the advisor is private. That boundary is what makes “you write privately, you share publicly” coherent.

Active and observer roles

Every participant has a role: active or observer. The asymmetry is a deliberate extension of the transparency contract, not a privilege gradient.

  • Active participants are the conversation’s writers. They send messages, their drafts run through an advisor, and the mediator processes what they send. Up to 3 active participants per conversation.
  • Observers bear witness. They can read every message and the full transparency stream — including the mediator’s reasoning and the active participants’ personas — but they cannot send messages, they have no advisor of their own, and they are never included in any LLM prompt. Up to 10 observers per conversation.

The asymmetry is always visible: an observer-count pill appears in the conversation header whenever observers are present, observer names are enumerated for screen readers, and observers see an “Observing” badge plus a “no composer” affordance so they always know their own relationship to the conversation.

Why not just mute some active participants? Because the LLM-visibility boundary is the load-bearing asymmetry. An “active but silent” participant would still feed into the mediator’s context; an observer does not. That maps onto how mediation, therapy, courtroom, classroom, and arbitration actually work — the structured-conversation parties are the ones being mediated between; everyone else is present but not party to the substance. Anonymous observers are explicitly out of scope: the roster pill exists so active participants can always see who is watching.

Distribution modes

The product is designed to eventually ship in two forms:

  • Hosted (paid): explAIn provides the LLM keys; users subscribe and consume from a token quota.
  • Self-hosted / BYO-key (open source): users supply their own provider keys; no billing.

The provider abstraction (packages/llm) is intentionally framework-free so it can lift into the future open-source variant unchanged.

Graph View

Backlinks