Technology
How Zaelar is built.
A layered architecture: a fast reflex layer for the moment-to-moment, live worker sessions for anything that takes real thinking, a memory that actually persists, and a canvas of small apps your agent builds and drives.
Why this shape
One agent, not a folder of apps.
Most software makes you be the integration layer — copying context between a dozen isolated tools. Zaelar inverts that: one agent, one memory, one place your digital life lives.
A different app for every task
One agent, one place, for everything
Every app kept its own data, on its own
One shared memory your agent draws on
Isolated tools, no memory between them
Context that carries across every task
The stack, piece by piece.
Architecture
The full system, end to end: voice, orchestration, memory and the executor layer that runs your agent's work.
Read more →FlashBrain
The reflex layer — a small, fast model that answers, controls apps, and decides what needs deeper work.
Read more →Brain Workers
Live, steerable sessions that carry out anything that takes real reasoning — coding, research, multi-step tasks.
Read more →Memory
A local-first store with hybrid retrieval — vector, keyword and graph search, reranked and reinforced by use.
Read more →Apps
On-canvas apps: use the built-in ones, fork and modify them, or create and share your own — your agent drives them all by voice.
Read more →Agent Network
How Zaelar talks to other agents over the MeshKore agent network — same brain, a stricter profile, and guards that only loosen with explicit consent.
Read more →Developers
Zaelar's core is public source — clone it, read it, run it on your own machine.
Read more →