Repository navigation
FAQ
Yes, except for AI requests. Startup, browsing history, search, prompts, settings, import/export, and data management all work without internet. Only the request to the selected AI provider needs the network. See Product-Overview.
In one local SQLite file (nexora.db) in the OS per-user app-data directory.
There are no accounts and no cloud sync; backup is a filesystem copy. See
Database.
Exclusively in the OS secure keyring — never in SQLite, logs, or source. SQLite holds only provider names and display labels. See Security-and-Credentials.
OpenAI, Anthropic, Google Gemini, xKiro, OpenRouter, NVIDIA NIM, and OpenCode Zen (last four via the OpenAI-compatible endpoint). Model lists are backend-maintained shortlists, plus a custom model ID field in Settings. See Providers-and-Models.
One user-started, bounded tool-use loop per dialog, with streamed step events and pause/resume/cancel plus an extendable iteration budget. No schedulers, no background daemons. See Agent-Runs.
supervised gates tool actions behind approval; semi_autonomous (default)
auto-allows routine actions under permission rules and parks sensitive ones;
full_autonomous has the widest latitude but keeps budgets, spend limits,
rules, and checkpoints. See Autonomy-and-Approvals.
The decision parks until you approve once, deny, or decide later. Parked approvals are never auto-resolved, including in the planned non-interactive CLI. See Autonomy-and-Approvals and CLI-2.1.
An optional per-run micro-USD budget; breaching it ends the run as
spend_limit_exceeded. Shown figures are limits/estimates, never guaranteed
cost. See Autonomy-and-Approvals.
Offline FTS5 over conversation titles, message content, and prompt titles plus contents. Results open the matching item. See Search-Import-Export.
Export writes one conversation to JSON without touching stored data; import inserts supported files as new rows with fresh IDs (never merges), atomically — an invalid file errors and rolls everything back. See Search-Import-Export.
Delete individual conversations or prompts directly, or clear all data behind an explicit typed confirmation. Deletes are hard deletes with cascades. See Security-and-Credentials.
No. CLI 2.1 is planned scope only (TUI + pipe-safe commands with the same safety floor) — not yet shipped. See CLI-2.1.