Sub-services

Internal layers that make Astronox feel safer and smarter

Trident MS, Peacock JL, and HookWare run under the main product surface to improve decision quality, memory continuity, and adaptation without extra setup.

Trident MSPeacock JLHookWareLocal-first

Jump To Layer

Trident MS Peacock JL HookWare

Trident MS

Trident Memory System

Trident is Astronox's post-session Auto-Dream layer. It captures durable context after each assistant turn and keeps memory continuity local to your device profile.

What It Does

Runs as a post-session hook after each assistant response is saved
Extracts durable signals from recent conversation context
Classifies memory into user, feedback, project, and reference buckets
Writes deduplicated entries into local memdir markdown files

Consolidation Modes

Heuristic mode (default) uses fast deterministic extraction
Model-assisted mode refines candidates through your active provider
If model-assisted output fails, Trident automatically falls back to heuristic extraction

Storage And Activation

Stored locally in memdir/MEMORY.md plus user.md, feedback.md, project.md, and reference.md
Includes memdir/.trident_state.json for runtime state handling
Enabled by default and configurable in Settings -> Sub-services -> Trident
Can be toggled from Supersearch with dedicated enable/disable and mode commands

Peacock JL

Peacock Layer

Peacock is Astronox's pre-execution judgment layer. It catches fragile intent before work begins and lets you decide whether to switch to a safer execution path.

What It Does

Analyzes requests for reliability and execution fit before normal flow starts
Surfaces safer alternatives for requests likely to fail or produce brittle outcomes
Pauses execution until you choose Accept, Reject, or Dismiss
Can surface one-click Undo for supported accepted suggestions

Common Intervention Patterns

Long timers that should be persistent scheduled reminders
Absolute time requests phrased as duration timers
Repetitive manual checks better handled by recurring automations
Manual click-path instructions that can be replaced by direct intent execution

Control Model

Accept applies the suggested action when reliability is higher
Reject continues your original request and can include feedback
Dismiss hides the prompt for that moment without setting a long-term preference
Under Max Reasoning or Coding Mode, high-complexity prompts can trigger parallel specialist fan-out

HookWare

HookWare is Astronox's local activity intelligence layer. It reads workflow rhythm at a pattern level to reduce prompt friction and improve long-term personalization.

Signal Categories

Active app usage and app-switching patterns
Media playback state signals
High-level terminal activity indicators such as command, warning, and error counts
Astronox workflow trends like panel usage and tool frequency
Broad web activity signals at site/title level plus runtime diagnostics

How It Helps

Tunes response depth and pacing to your observed operating style
Prioritizes more relevant suggestions during active workflows
Reduces noisy interruptions during heavy focus windows
Improves provider and model preference alignment over time

Consent And Controls

Opt-in during onboarding and user-controlled from Settings
Category-level toggles let you keep only useful signal sources
User model profile integration allows profile summary checks in chat
Best evaluated after several days of non-idle usage rather than first run

Want the broader architecture view beyond these internal layers?

SEE CAPABILITIES