What release 1 does not include

The plain list of what the first release of agents in Osy# does not do: no starter app, no editing of limits on a screen, no screens that update by themselves, no run-on-demand button, no process drawn on a canvas, no tests made from corrections, teaching that is for all of an app's agents at once, and a few more. Release 1 starts careful on purpose: an agent does little alone, and few people can change what it does.

You are here: Agents in an Osy# app › What release 1 does not include

Summary#

Release 1 is deliberately careful. Said nothing, an [[agent-guide-glossary#agent|agent]] only looks things up. Few people can widen what it does. Everything it does is written down. The list below is what was left for later, so that nobody has to find out by trying.

Everything else in this guide describes what is built and tested.

Description#

Getting started#

No starter app with agents in it. There is no ready-made project to begin from, and no sample app to copy with osy docs sample. The smallest working apps are on the developer pages, whole: Shape 1 — an assistant for one person and the three after it.
The kits are not yet published as packages. They are in the Osy# source tree and tested there.

Running agents#

Limits and settings are not edited on a screen. The management screens show every limit and what has been spent against it. A limit is changed in the app's code, or by a function a developer calls.
No "run it now" button. Work is given to an agent by the app, not from the agent's page. An agent that has never worked says so.
An agent is not retired from a screen. It is paused there; removing it is a change to the app.
Screens do not update by themselves. Each shows what is true when it is opened, and has a way to refresh.
Abandoned work is not closed on a timer. The function that closes work whose process died exists; an app schedules it, or a person in charge calls the work off.
Nothing is announced when a process run gets stuck. The list of runs shows which are stuck; no event is raised for the app to act on.

Agents working together#

Handing work on is the [[agent-guide-glossary#model|AI model]]'s choice among what the app declared. An app's own code can start several agents at the same time, as separate pieces of work. It cannot yet direct one coordinator step by step, with the parts gathered under one receipt and one limit.
No ready-made patterns for groups of agents, such as a debate or a review round.

Teaching#

Teaching is for all of an app's agents at once. Whoever teaches, teaches every agent and reads every agent's work with what it said. It cannot be given for one agent only.
A serious [[agent-guide-glossary#correction|correction]] does not, by itself, reduce what an agent may do alone. A person in charge does that.
Changing an agent's [[agent-guide-glossary#proposal|proposal]] before approving it is not recorded as a correction. The person rejects it with a reason, or approves it as it stands.
No test cases made from corrections, and no agent that reviews its own past work on a schedule.

Processes#

A process is written, not drawn. There is a picture of every process and of every run, drawn by the app. There is no canvas on which a person assembles a process from agents.

People and requests#

An agent cannot ask somebody outside for a document. It can ask a question, ask for an approval, and ask for the fields of a form. Receiving a file from a person with no account is not part of it.
"Refunds go to Finance" holds only when Finance's [[agent-guide-glossary#inbox|inbox]] is restricted to its members. An inbox that is not restricted is worked by everybody who works the app's inbox.

Privacy#

A few names are outside a person's privacy request: the name on a standing permission they gave an agent, on a hand-over of work, on a pause or a change of [[agent-guide-glossary#sponsor|sponsor]], and on a question they were asked. What agents did for them and about them is covered.

A known fault#

On the management screens, inside the app frame that has a side bar, an error notice can appear at the foot of some screens in a browser. The screens draw and work.

Go deeper#

[[agent-delegation#limits]] — what handing work on does not do yet
The management screens — where a person runs an app's agents (Agents.Management kit) — what the management screens do
Correcting an agent, and what it is told from then on — corrections, instructions, instruction sets and rollback (Agents kit) — corrections and instructions