Decisions¶

Labels: CONFIRMED = stated as decided; PROPOSED / PLANNED = not yet decided; SUPERSEDED = replaced by a later decision (the replacement is named). Every line cites its source.

Decisions from the owner's source of truth (v4, 2026-10-23)¶

Source: the owner's source-of-truth document, version 4 (2026-10-23), §10. These outrank discussion decisions.

CONFIRMED The harness is open source: the RSI engine and the agent library with an empty pattern slot ("bring your own pattern"). Our tuned pattern stays private.
CONFIRMED The guard and the memory substrate stay inside the hosted service and arrive together on connection; neither is offered alone.
CONFIRMED Stages: pre-alpha (friends and family, hosted end to end, guard included), then alpha, beta, gamma.
CONFIRMED Pricing is metered on memory space, operations and concurrent agents; the customer's own model subscription pays for compute. Not per token. Supersedes the token-economy / 15% capability-pool model on the website and in the May 2026 notebook.
CONFIRMED Cross-model drift realignment is a real capability to build and measure; it joins the pre-alpha hand-off test.
CONFIRMED Public line: "your AI's memory belongs to you, not the model provider."
CONFIRMED Public wording: the mechanism is called cessation; the engine's internal name and doctrine terms do not appear in public copy (§ "Words in public").
CONFIRMED Nothing from the harness goes public before the guard freeze score is signed (§7).

Earlier decisions recorded in the CLI's bundled docs¶

Source: docs/decisions/ADR-001..008 in the installed CLI v0.2.9 (see cli-investigation). Dates not stated in the snapshot.

ADR-001 Gemini Live for Sinhala voice. Status now: UNRESOLVED (voice is "specced" on the site).
ADR-002 OpenRouter only. SUPERSEDED by four BYOM backends (site) and claude-cli use (June 2026 test sessions).
ADR-003 Firebase realtime DB; ADR-004 edge-first voice; ADR-005 two-stage Vinaya; ADR-006 swarm experiment loop; ADR-007 sanctuary meta-skill; ADR-008 [doctrine term] operational framework. Status vs source of truth v4: UNRESOLVED.

Decisions from the discussions¶

Newest first. Summarised automatically from the discussions; not reviewed by participants.

28 Sep 2026 ¶

Decided · Speaker 1 will serve as the first testing customer for the harness prototype.
Decided · The team will not build a fixed UI upfront, but rather generate dynamic on-the-spot interfaces based on prompt specifications.

23 Feb 2026 ¶

Decided · The first milestone is validating the current demo website container before making it publicly accessible.

9 Feb 2026 ¶

Decided · Use Telegram instead of WhatsApp for the bot interface due to official API support and cost efficiency.

2 Feb 2026 ¶

Decided · User interactions, common patterns, and structured concept states will be persisted in a vector database to enable personalized agent memory.

19 Jan 2026 ¶

Decided · Use Discord server as the primary messaging integration over MS Teams for office communication testing.

12 Jan 2026 ¶

Decided · Maintain original song lyrics without modification when experimenting with modern remixes or EDM styles.

29 Dec 2025 ¶

Decided · Remove the separate breakdown for Phassa, Vedana, Sanna, and Cetana from the Citta-Cetasika analysis section and navigate directly into the analysis.
Decided · Disable and hide the Dhamma Upakaraka feature for now as it is not currently in use.
Decided · Remove the non-functioning graph and analysis JSON boxes in the comparison view and keep the working table comparison.
Decided · Move the Citta search/table section further down beneath the Cetasika analysis section and remove the 'Trividyawa' section.

15 Dec 2025 ¶

Decided · On machine.io landing page, Inner Realm and Marga Sakaccha links will be made green and active, while Machine World will show 'Launching Soon'.

13 Oct 2025 ¶

Decided · Keep the backend light and do not build custom front-ends or back-ends beyond Firebase.

28 Jul 2025 ¶

Decided · The team agreed to conduct weekly catch-up calls on Thursdays.