Skip to content

Latest commit

 

History

History
187 lines (148 loc) · 10.6 KB

File metadata and controls

187 lines (148 loc) · 10.6 KB

LISA docs

Everything in docs/ — indexed, so you can tell in one screen whether a file is something to read, something to implement, or something that already happened.

The rule this index enforces: a plan whose work has shipped moves to archive/plans/. What stays in docs/ is either documentation you use or work still open. If a plan here is finished, archive it.


Start here

Doc What it is
GUIDE.md · 中文 The full user guide — install, every surface, soul, knowledge base, mail, channels, heartbeat, tools, config, REPL, layout. The README is the 60-second version.
PROVIDERS.md Ready-to-use configs for 20+ LLM providers, including local ones.
CODING_PLANS.md Running Lisa's coding work on a Claude Pro/Max, ChatGPT or Copilot subscription instead of a metered key — and why token extraction was refused.
PTY_AGENTS.md Steering a real claude / codex CLI through a pseudo-terminal, and adopting a session you started yourself.
RELEASING.md How to cut a release: version bump, notes, changelog, tags, npm, Homebrew, signing and notarization.
PUBLISH.md The distribution channels themselves and the one-time credentials each needs.
CONTRIBUTING.md Where to help, the norms, the local dev loop.

Current plans

Open work. Each plan carries its own status line — trust that over this table.

The spine

Plan Scope
ROADMAP_v1.0.md The 1.0 roadmap: four pillars (Sense · Dispatch · Reve · Model) and what "1.0" means. Never archived — it is the map.
AUTONOMY_ROADMAP.md The autonomy arc (idle, heartbeat, examen, soul evolution) behind the Reve pillar.

Four pillars

Plan Scope
PLAN_SENSE_v1.0.md Resident ambient perception, consent-gated.
PLAN_SENSE_S2_v1.0.md Ambient vision + voice — the most privacy-sensitive block; gated on the consent framework.
PLAN_SENSE_SOCIAL_CONNECTORS_v1.0.md Conversational social publishing through connected accounts.
PLAN_DISPATCH_v1.0.md The unified command layer over local CLI agents + TakoAPI.
PLAN_DISPATCH_D2B_v1.0.md D2b — an A2A adapter putting remote agents in the hub.
PLAN_DISPATCH_D4_v1.0.md D4 — the multi-agent monitor and advisor actions.
PLAN_OBSERVER_DEEPENING_v1.0.md Tier-2 fidelity for the codex / opencode / aider observers.
PLAN_REVE_v1.0.md Hardening autonomous reflection: quality gates, budgets, observability.
PLAN_MODEL_v1.0.md From "bring your own endpoint" to real local deployment.
PLAN_FOUNDATIONS_v1.0.md The cross-cutting floor the other three stand on: consent, security, tests.

Cloud, accounts, identity

Plan Scope
PLAN_CLOUD_v1.0.md LISA Cloud on GCP. M0 deployed; per-user isolation still open.
PLAN_ACCOUNTS_BILLING_v1.0.md Centralized accounts and token billing.
PLAN_IDENTITY_v1.0.md "One identity, two data planes" — who you are vs. where your Lisa lives.
PLAN_WEB_SIGNUP_v1.0.md Signup / login on meetlisa.ai and the signed-in web app.

Capabilities

Plan Scope
PLAN_HARNESS_ALIGNMENT_v1.0.md Aligning with the DeepSeek harness: capability seam, sandbox modes, prompt in the session log.
PLAN_KNOWLEDGE_BASE_v2.0.md Knowledge base v2: URL ingest, the daily brief, the link graph.
PLAN_DESIRE_EVOLUTION_v1.0.md Why her wishes didn't change with a conversation, and how to fix it without breaking sovereignty.
PLAN_MAIL_v1.0.md The read-only mailbox and its daily digest.
PLAN_MODEL_TUNING_v1.0.md Per-model knobs worth keeping for the Anthropic provider.

Clients and surfaces

Plan Scope
IOS_COMPANION_PLAN.md The Lisa Pocket design.
PLAN_IOS_ONBOARDING_v1.0.md First launch and pairing on iOS.
PLAN_IOS_REACHABILITY_v1.0.md "Always reach Lisa" — R1–R4 landed, R5 waits on cloud accounts.
PLAN_MAC_APP_STORE_v1.0.md Whether Lisa.app can ship on the Mac App Store (verdict: only the thin client).
MAC_ISLAND_PLAN.md The menu-bar island.
PLAN_ROOM_v1.0.md The room she lives in — research and design. Phase 1 landed; v2.0 is archived.
PLAN_GRAMOPHONE_v1.0.md Opt-in ambient music in the room.
PLAN_WEBSITE_REBUILD_v1.0.md Rebuilding meetlisa.ai in the Hakko design language.

Older plans, kept for their reasoning

Plan Scope
ORCHESTRATOR_PLAN.md The original "LISA as mission control" plan (2026-05).
PRODUCTIZATION_PLAN.md CLI polish / website / providers / Mac app under the sovereign-only, OSS-only constraint.
SPRINT_4_PLAN.md A data-gated candidate list, deliberately never committed to.

Completed plans (archive)

Shipped. Kept because the reasoning and the debates are still worth reading — several are cited by name from source comments.

Plan Shipped as
PLAN_AUTH_OTP_GOOGLE_v1.0.md Email-OTP + Google Sign-In (status: IMPLEMENTED, 2026-07-24).
PLAN_KNOWLEDGE_BASE_v1.0.md The 3-layer personal knowledge base (PRs #235–#242).
HANDOFF_KNOWLEDGE_BASE_v2.0.md The K-series handoff; the work landed in v0.21.0.
PLAN_UI_SESSION_SHELL_v1.0.md The three-column session shell, phases 1–3 (PRs #343–#345).
PLAN_UI_SESSION_SHELL_v1.1.md Its follow-up wave F1–F7 (PRs #346–#350).
PLAN_ROOM_v2.0.md The room's life system, phases A–E (PRs #221, #222 + 换景).
PLAN_DESIRE_DYNAMICS_v2.0.md Desire dynamics v2, PRs A–C including cloud parity.

Releases

RELEASING.md is the how. CHANGELOG.md is generated from the notes below by npm run changelog — so the note is the source, and the changelog is the summary.

v0.24.0 · v0.23.0 · v0.22.0 · v0.21.0 · v0.20.0 · v0.19.0 · v0.18.1 · v0.18.0 · v0.17.0 · v0.16.0 · v0.15.0 · v0.14.0 · v0.13.0 · v0.12.0 · v0.11.1 · v0.11.0 · v0.10.0 · v0.9.1 · v0.6.0 · v0.5.0 · v0.4.0 · v0.3.1 · v0.3.0 · v0.2.0

Reviews

Point-in-time assessments. Each names its baseline commit; read the newest first.

Review Baseline
PROJECT_REVIEW_AND_OPTIMIZATION_v0.21.0.md Whole-project review, 2026-07-26.
REVIEW_IOS_APP_v1.0.md Lisa Pocket layout + functionality.
PRODUCT_REVIEW_v0.9.md Product capability review at v0.9.
PRODUCT_REVIEW_v0.3.md The first capability review + tuning plan.

The current pair — PROJECT_REVIEW_UX_v0.24.0.md and PROJECT_REVIEW_TECH_v0.24.0.md (2026-09-05, baseline 26266a5) — are the newest and supersede the v0.21.0 review.

Runbooks

Operational procedures for things that are live.

Runbook For
RUNBOOK_CLOUD_PROD.md The production Cloud Run deployment.
RUNBOOK_ACCOUNTS_LAUNCH.md Launching accounts / auth: the operational config the code alone doesn't cover.
WEBSITE_OPS.md meetlisa.ai — deploys, DNS, rollbacks.
FOOTPRINT.md What the resident service actually costs in CPU, memory and wakeups.
OBSERVER_FIDELITY.md Which fields each observer really produces, with the live-verification log.

Research

Background reading that shaped a design, kept separate from the plans so a finding is never mistaken for a commitment.

Note Subject
RESEARCH_DEEPSEEK_HARNESS.md The dsh harness, and what LISA should take from it.
RESEARCH_KNOWLEDGE_INTERNALIZATION.md How an agent might internalize what it reads.
RESEARCH_LEARNING_IN_REFERENCING.md Learning-in-referencing — the research track behind Reve.
RESEARCH_OPEN_SOURCE_GAME_MODELS.md Open-source game/character models surveyed for the room.

Design notes

Smaller, self-contained design write-ups that aren't plans with phases.

Note Subject
DESIGN_COMPRESSION_GATE.md The compression-gain consolidation gate.
DESIGN_CONCEPT_BENCH.md A constructed-language benchmark that turns misunderstanding into a controlled variable.
DESIGN_LEXICAL_ENTRY.md Token-triggered lexical-entry memory.
SHOW_HN_DRAFT.md The launch post draft.