General
Sidebar behavior, agent execution mode, data path, and the agent's persona files.
The General tab controls how the sidebar and agent behave, and lets you shape the agent's persona.
Sidebar & Agent
| Setting | Options | What it does |
|---|---|---|
| Sidebar default | Auto-open / Manual only | Whether the sidebar opens automatically |
| Agent execution | four tiers (see below) | How much the agent may do without asking |
| Data storage | path | Where artifact and session data are stored (e.g. ~/.config/nevoflux/) |
These save instantly.
Agent execution tiers
New in v0.3.13, Agent execution is a four-step ladder rather than a confirm/auto toggle. Each tier auto-approves everything the tier below it does, plus one more class of tool. Anything a tier does not cover still raises a permission prompt.
| Tier | Adds | Still asks before |
|---|---|---|
| Read-only (default) | Reading a page, navigating, waiting; web search + fetch; knowledge-base reads; thinking, planning, and telling you something | Everything else |
| Browser auto | + browser interaction: click, type, fill, scroll, key presses | Local files, shell, everything else |
| Browser auto + reads | + local file reads (read_file, list_files, glob, grep) | File writes, shell, everything else |
| Full auto | Everything — no confirmations | Nothing |
Two things that are easy to guess wrong:
- Navigation counts as read-only. Opening a URL is in the bottom tier; interacting with what loads is not.
browser_eval_jsis not a browser interaction. Running JavaScript in a page is in the top class with shell and file writes, so it needs Full auto — it can do far more than a click.
Tools the classifier doesn't recognize — including new, MCP and third-party ones — always fall into that same highest-risk class, so an unknown tool is never auto-approved below Full auto.
Upgrading from before v0.3.13? The old Auto-execute option was a stub that never actually took effect. It is deliberately not migrated to Full auto — it resets to Read-only, because silently granting full permissions during an upgrade would be worse than asking you to re-pick. Check this setting after updating.
The tier also bounds what a session can create: see loops and execution tiers.
Persona ("soul") files
Five Markdown cards let you shape who the agent is and what it knows about you and your environment. Each card is saved to a file in your config directory when you click Save All Settings:
| Card | File | Use it for |
|---|---|---|
| Identity | IDENTITY.md | The agent's name, character, vibe (and an Avatar) |
| Soul | SOUL.md | Core values and behavioral principles |
| User | USER.md | Facts about you (name, pronouns, timezone, context) |
| Tools | TOOLS.md | Your environment (SSH hosts, devices, TTS preferences) |
| Agents Runtime | AGENTS.md | Runtime instructions, memory rules, safety protocols |
Example: paste a short bio and your timezone into User, then Save All Settings — the agent will tailor answers to you. Set a name and personality in Identity to give the assistant a consistent voice.
These files also participate in NevoFlux's self-learning system, which can propose updates to them over time.
Per-Space personas (Space Souls)
New in v0.3.12, you can give each browser Space its own assistant persona — a "research" soul in one Space, an "engineer" soul in another — from the Space Souls settings. A soul, its private memory, and that Space's cookie jar all share one key. For the concept and why you'd use it, see Features → Space Souls.
Bindings are stored in space_souls.toml in your config directory, mapping a Space's
container id to a soul (role) slug:
# ~/.config/nevoflux/space_souls.toml
[bindings]
"firefox-container-1" = "research"
"firefox-container-2" = "engineer"If a Space isn't bound (or the file is absent), the assistant behaves exactly as it does with the single global persona above.