Skip to content

Product Overview

Jump to bottom
sshdw edited this page Oct 15, 2026 · 1 revision

Product Overview

Nexora is a local-first AI desktop workspace: chat with multiple AI providers, organize conversations, reuse prompts, and attach local documents — all stored in a single local SQLite file (nexora.db) in the OS per-user app-data directory.

MVP scope (SRS FR-001–FR-015)

  • Startup and offline access: launches without internet; local data, history, and settings stay usable offline (FR-001, FR-015).
  • Conversations: create unlimited conversations; rename, archive/restore, delete; full history persists across restarts in chronological order (FR-002, FR-005, FR-006).
  • Messages: user and AI messages visually distinguishable; failed requests show an error and are not persisted as messages (FR-003).
  • Providers and models: select provider and model before sending; selection applies to subsequent requests (FR-004).
  • Prompt library: create, edit, delete, insert into conversations (FR-007).
  • Documents: attach local files, review before sending, remove before sending (FR-008).
  • Search: offline search over conversations and prompts (FR-009). See Search-Import-Export.
  • Import/export: export individual conversations; import supported files with atomic rollback on validation failure (FR-010, FR-011).
  • Settings: persisted appearance, provider/model defaults, conversation, export, and data-management preferences; invalid values rejected (FR-012).
  • Data management: delete conversations/prompts; clear all data behind explicit confirmation (FR-013).
  • Credentials: add, update, remove provider keys; missing credentials detected before sending (FR-014).

Shipped agent stack (beyond MVP)

  • One agent run per dialog, with streaming step events.
  • Tools: file read/write, command execution, directory listing.
  • Autonomy ladder supervised / semi_autonomous (default) / full_autonomous, approval parking, pause/resume/cancel, iteration budget with extend, per-run spend guard. See Agent-Runs and Autonomy-and-Approvals.
  • Classified provider errors (rate limit with Retry-After hint, outage/5xx, network/timeout, rejected credential, invalid request, unexpected response) for chat and agent runs alike.

Non-goals (explicitly out)

No cloud sync, user accounts, shared workspaces, collaboration, mobile/web apps, plugin ecosystem, voice interaction, or unattended automation (no schedulers or background daemons). User-initiated agent runs and bounded task loops are shipped and are not part of this exclusion list.

See also: Architecture, Roadmap, FAQ.

Clone this wiki locally