An agent session is a chat with Elzi inside a project folder. You say what you need, and the agent plans, works with the tools you've allowed and asks when it needs you. It only works inside that folder. One model, node or strategy answers the chat, and you can change it at any time. When you want an agent that keeps working without you starting each turn, Agent Work runs it on a schedule.
Chat sessions
- Under Projects, pick a folder and choose "New session here". Any folder becomes a project the moment a session runs there.
- Pick the model with "Run this chat on": a node, a strategy, or "Other models and providers…".
- Type what you need in plain words and press Enter.
Sessions are listed under the folder they ran in. The sidebar shows which ones are active now and which are waiting for your answer. You can fork a session to try another direction, or delete it. When an agent hands part of the job to delegates, they share the chat's nodes. Skills, MCPs and Memories extend what the agent knows and which tools it has.
Agent Work: agents on a schedule
Agent Work is for an agent that keeps a desk: it wakes on a schedule, checks what's new, investigates, keeps a ledger and a scratchpad, drafts, has its drafts reviewed and asks you only when a decision is yours. Each wake-up is a tick. An agent is a folder with an agent.yaml file.
Choose "New agent" and fill in its purpose and folder, a charter (who it speaks as, what it may touch, what it must ask about first), a schedule (how often it wakes, on which days and hours), its tools and access, a guard that runs first and can hold a tick, specialist workers and review lenses that each attack every draft. "Open desk" shows what needs you, the current tick's lanes, a ledger and the tick history. "Remove from roster" keeps the folder.
Good to know
Agent Work and the Benchmark view appear in the default local edition of the app.
How it differs from builds and benchmarks
| Agent session | Software build | Benchmark run | |
|---|---|---|---|
| You start it with | A chat in a project folder | A chat on "Elzi · Build" | The Benchmark view |
| Who drives | You, turn by turn (or a schedule, for Agent Work) | The swarm, from one brief | A frozen task, the same for every model |
| Models | One model or node answers | Many nodes share the work | The model or nodes you want to measure |
| What you get | The conversation and the files it changed | A planned, multi-agent build written into the project | A score with its evidence, which you can compare |
Facts on this page were checked against our own test records on 6 November 2026.