Skip to content

README.md

Latest commit

 

History

History
46 lines (35 loc) · 6.13 KB

README.md

File metadata and controls

46 lines (35 loc) · 6.13 KB

Alice vNext Public Preview

Alice vNext is the next public wedge for Alice as an agent-native second brain: a local-first memory kernel, governed explicit agent memory commits, reviewable generated briefs, model-backed source-grounded synthesis, live local connector-backed evidence capture, hardened connector settings/state/secrets, agent-facing context packs, governed agent proposals, and a local scheduler runtime.

This preview is not a hosted launch. It is a repo-local, deterministic public preview built around the vNext memory-kernel schema and the fixture-safe workflows shipped under v0.5.1-vnext-preview.

Product Shape

Alice vNext has three layers:

  • Alice Core: the local-first persistence, provenance, policy, and event-log substrate. Core owns sources, chunks, memories, revisions, graph edges, projects, open loops, artifacts, evals, and connector evidence.
  • Alice Brain: the user-facing second-brain workflows on top of Core. Brain generates daily briefs, weekly syntheses, context packs, contradiction reports, connection reports, project updates, open-loop reviews, and reviewable artifacts in deterministic or model-backed mode.
  • Alice Agent Memory: the agent integration layer. Agent Memory exposes continuity through CLI, API, and MCP so external agents can capture, retrieve, resume, explain, generate context, explicitly commit user-directed memory through policy, propose reviewable memory, and trigger governed scheduler workflows without owning the memory database.

Preview Surfaces

  • Source capture: manual text, local text/Markdown files, Markdown folders, ChatGPT exports.
  • Retrieval: deterministic context packs with domain/sensitivity filters and provenance.
  • Brain workflows: daily brief, weekly synthesis, connection report, contradiction report, project update, open-loop review.
  • Model-backed intelligence: provider/routing abstraction, local-first model policy, source-grounded sections, prompt hashes, context hashes, model metadata, and deterministic-vs-model comparison mode.
  • Quality review: artifact ratings for usefulness, accuracy, source grounding, novel connections, actionability, hallucination risk, verbosity, missed context, and comments.
  • Agentic control plane: scoped agent identities, permission profiles, policy decisions, explicit trusted memory commits, inline confirmations, memory proposals, undo/correction/forget lifecycle controls, and Agent Activity audit surface.
  • Governed scheduler: disabled-by-default workflow controls, a local daemon runner, due scans, run history, trace IDs, failures, duplicate-run locks, and reviewable artifacts.
  • Connectors: allowlisted Telegram sync, local folder/Obsidian scan and watch, browser clipper capture endpoint, Hermes/OpenClaw-style agent output ingestion, dedicated settings/state rows, local encrypted secret references, retry/cursor hardening, plus deterministic PDF, DOCX, CSV, screenshot OCR, and voice transcript payload ingestion.
  • UI: live/fixture-backed /vnext workspace for source review, source archive, capture-to-brief traces, Ask Alice, briefs, queue, projects, Agent Activity, trusted memory commit audit, inline confirmations, Schedules, beliefs, graph, live connector configuration, connector health/defaults/bookmarklet guidance, dogfooding readiness telemetry, doctor/readiness checks, privacy settings, model comparison, and quality ratings.
  • Evals: synthetic corpus and baseline metrics for recall, temporal reasoning, contradictions, provenance, privacy, open loops, and prompt injection.

Start Here

  1. Follow public alpha quickstart for the design-partner install path.
  2. Use first-run checklist and doctor for onboarding.
  3. Review headless Ubuntu install, Hermes dogfood on Ubuntu, agent integration pack, MCP tools, Hermes skill, and OpenClaw skill.
  4. Follow vNext quickstart for the broader preview path.
  5. Review architecture.
  6. Review security and privacy and public alpha security posture.
  7. Review local runtime before running scheduler workflows in the background.
  8. Use example ALICE.md as the first Brain Charter.
  9. Use demo script and demo mode for a short walkthrough.
  10. Use release checklist before publishing or tagging.
  11. Review preview release notes, public alpha release notes, and tag plan.
  12. Review the dogfood daily checklist before daily local-alpha use.
  13. Review the agentic control plane CTO summary, agentic memory commit CTO summary, local runtime CTO summary, model-backed intelligence CTO summary, live capture connectors CTO summary, dogfood hardening CTO summary, live-backed operator console CTO summary, public alpha packaging CTO summary, and headless Ubuntu packaging CTO summary for sprint closeouts.

Launch Boundary

The public preview should prove that a technical user can install Alice locally, capture live local evidence, configure local connector defaults safely, run readiness checks, and generate a first daily brief in under 20 minutes. It should not claim managed connector OAuth, packaged browser extensions, hosted connector polling, cloud sync, hosted SLA, or automatic promotion of generated artifacts into trusted memory.