Context is the missing primitive of AI.
Maple was built with a core thesis: the limitation of modern AI isn't the model's intelligence—it is the model's sight. We eliminate the copy-paste ritual.
The Copy-Paste Friction
Traditional chatbots force you to act as a data integration pipeline. Every query requires manual preparation.
- Context Sifting: Finding, opening, and copy-pasting code fragments or conversation histories just to explain a bug.
- No Memory Continuity: A new chat window is a complete wipe. You must re-introduce your stack, conventions, and objectives.
- Call Amnesia: Decisions made during Zoom or Meet calls stay locked in transcripts, away from where you actually write code.
The Maple Flow
Maple connects directly to your desktop environments and tools, pulling references automatically.
- Screen-Aware Prompts: Maple reads your active window automatically. Summons take less than 100ms.
- Unified Context Sync: Integrations with Slack, Notion, and local file storage are indexed locally and linked in memory.
- Post-Meeting Synthesis: Structured meeting actions are saved automatically, immediately queryable in Focus Mode.
Under the hood of the Context Engine
Local-first, secure, and fast. Maple turns your desktop activity into a searchable graph database without compromising your system resource limits.
Local Vector DB Cache
Uses SQLite with Write-Ahead Logging (WAL) caching. Keeps searches running under 100ms by running index compaction tasks during system idle states.
Zero-Knowledge Privacy
All data indexing, optical character recognition (OCR), and vector storage happen locally. Raw text never leaves your device; only relevant text chunks are sent to LLMs during queries.
Resource Optimization
Built with native Rust layers to monitor CPU temperature and battery states. Stops heavier background indexing processes when running on battery power.
How Maple Compares
A capability matrix comparing Maple with generic AI interfaces and single-purpose recording bots.
| Capability | Maple | ChatGPT / Claude | Meeting Recording Bots |
|---|---|---|---|
| Screen Awareness | Real-time OCR | Manual file upload | None |
| Zero-Paste Operation | Yes (⌥ Space) | No | No |
| Meeting Synthesis | Transcribes & Auto-tags | Paste manual text | Transcribes only |
| Local Index Encryption | Yes (AES-256) | No (Cloud storage) | No (Cloud storage) |
| Schedule-Based Digests | Custom Routines | None | None |
Technical Specifications & FAQ
Answers to architectural, integration, and security questions from developers.
Does indexing local files run down my laptop's battery?
No. Maple's local crawler runs as a low-priority background thread in Rust. It monitors hardware stats and pauses indexing whenever battery saver is active, or if CPU utilization exceeds 65%. On-device embedding runs selectively.
How does meeting transcription work without bot invitations?
Maple captures system-level audio output and microphone input securely when a virtual meeting window is active on screen. It does not require a bot listener to join the call, keeping your calendar invite list clean.
Which third-party models are used?
By default, Maple utilizes Gemini 1.5 Flash or Claude 3.5 Sonnet for cloud inference requests, sending strictly filtered context tokens. Q4 2026 scheduling targets native offline operations using Llama 3 models on hardware with NPU or dedicated GPUs.
