Chat tiles
Every tile in the Salana App's Talk view is an agent session running in the folder. Several side by side, each on its own task, each with the whole memory to read.
Product · Agents
An agent is a model with hands, working inside your memory.
An agent is an AI model given tools: it can read your files, run your skills, search the web, call your apps and write outcomes back. In a Salana Memory every agent works inside the folder, so it starts with everything you know.
What an agent is
A chat answers. An agent does. It reads the relevant part of your memory, decides what the task needs, runs the steps (a script, a search, an API call, a file edit) and records the outcome where the next agent will find it.
Today Claude Code and Codex work as agents in the memory, shown as tiles in the Salana App. Any command-line agent can be added as an engine; the memory does not care which model did the work.
How agents are used
Automations
An agent runs on a schedule or a trigger, with the skill it needs and a definition of what a good run leaves behind.
Change model
Start in Claude, continue in Codex with a click and carry on. Or let the two debate a document until they agree on a version.
Voice
Say the wake word, ask, and the agent answers aloud. It knows the names in your vault, so it hears them right.
Pages
Ask for a dashboard or a board and the agent builds it as a page in the Salana App, from the files in your memory.
Task logging
The agent opens a card when it starts substantial work and closes it with the outcome. The board writes itself.
The rules the agents follow
Read by default, write when confirmed
A chat runs with read-only tools until you confirm a write for that turn.
The conventions are in the folder
How the memory is organised is written inside it. Every agent reads it first, whichever model it is.
Outcomes are graded
An automation says what a good run leaves behind, and it is checked after every run. A red run alerts you.
One memory, not an org chart
We do not sell you a marketing agent, a sales agent and an HR agent with separate memories. One memory, shared. Agents come and go per task; the memory stays. Any model, any task, same context.
Set up a Salana Memory that is specifically yours.
Your files, every model, yours forever.