A local-first desktop tool with native integrations. No browser wrappers. No data harvesting. Your keys, your machine, your rules.
Sixteen typed memory cells arranged across persistence tiers and bleed axes. Volatile scratch pads evaporate on restart. Session echoes fade after 7 days. Durable workspaces hold for 30. Permanent bedrock and lore are encrypted at rest and survive forever.
The bleed axis determines how memory propagates. Ambient bleeds everywhere. Reactive bleeds only to direct follow-ups. Indexed bleeds through semantic search. Archive bleeds never — it is write-only recall. You can override any cell's behavior through the settings panel or mod cartridges.
Five capture sources without leaving your desktop: full screen, active window, custom region, webcam, and file upload. OCR runs locally. The accessibility tree gives Aerenium structural context for interfaces, not just pixels.
Vision analysis routes through your chosen model (GPT-4o-mini, Claude, or local). The accessibility tree traverses up to 150 elements across 4 levels. Vision-guided targeting uses fuzzy bounding-box search to locate UI elements even when coordinates shift. You can define custom capture regions that persist across sessions.
Aerenium does not teleport the cursor. It moves along Bezier curves with 4 px of human jitter, 8% overshoot on target acquisition, and pauses proportionally to distance. Keystrokes are delivered as typewriter input with 45 ms intervals.
The automation engine respects your desktop. It checks window focus before acting, avoids inputs during fullscreen video, and pauses when you move the mouse manually. The .ndjson audit log records every action with timestamps, coordinates, and screenshots for review. You can replay any session from the log.
Join voice channels as a native bot. Capture system audio via WASAPI or PulseAudio at 48 kHz stereo PCM. Control Spotify playlists, queue tracks, and mix voice conversation with music playback.
The Discord bot supports text commands, slash commands, and voice activity detection. Music control works with Spotify Connect and local audio files. System audio capture is opt-in per session and shows a persistent indicator while active. All audio stays local — no cloud transcription or processing.
Spawn parallel workers with scoped objectives. A mission system tracks progress across multi-step workflows. Buddy mode keeps the interface conversational and present. Study mode focuses on research and synthesis with reduced chatter.
Sub-agents inherit the parent memory context but run in isolated sandboxes. Missions can branch, retry failed steps, and request human approval at checkpoints. Buddy mode maintains continuity across sessions. Study mode disables non-essential chatter and focuses on document analysis and synthesis. You can define custom modes through mod cartridges.
Your API keys never leave your machine. The local cache is encrypted. License tokens are hardware-bound and re-verified every 5 days while online, with graceful offline degradation if the network drops.
The hardware ID is a SHA-256 hash of system identifiers, truncated to 16 hex characters. It cannot be reversed to reveal your identity. License tokens are RS256-signed JWTs with expiration. The local cache uses Fernet encryption keyed to your hardware ID. Enterprise customers can run the license server on-premise for complete air-gapped operation.
Aerenium exposes a broad surface of capabilities you can compose into workflows. From file management and browser automation to audio processing and system integration.
