Skip to content
LykeionDocs

Search documentation

Documentation/Get started

Core concepts

Before assigning research, separate the question you want answered from the software that executes it. You need only a signed-in workspace to browse these objects; executing work also needs a machine and a usable agent runtime.

EXPLORE THE CONNECTIONS

A shared place for research.

Select a concept to trace its relationships. Use Tab to enter the map and arrow keys to explore.

Choose a concept

Workspace

The shared context for studies, colleagues and configuration.

Connected concepts: Study · Agent · Team · Skills · Connectors · AristotleRead Workspaces

Study

A research question or task with an assignee and configured stage scope.

Connected concepts: Workspace · Stages · Agent · Team · Run · Evidence · AristotleRead Studies

Stages

Grounding, Hypotheses, Experimentation and Validation describe the process. A study can cover part of that sequence.

Connected concepts: Study · Agent · Team · EvidenceRead Scientific stages

Agent

A reusable research role with instructions, capabilities and execution settings. Each run is an individual attempt.

Connected concepts: Workspace · Study · Stages · Team · Machine · Run · Skills · ConnectorsRead Agents

Team

Coordinated research roles around a stage, with members, instructions and review criteria to inspect.

Connected concepts: Workspace · Study · Stages · Agent · SkillsRead Teams

Machine

A computer exposing agent CLI runtimes. Check the runtime is online and healthy before assigning execution.

Connected concepts: Agent · RunRead Connect a machine

Run

One execution attempt, with its own progress and errors. An agent profile is reusable across attempts.

Connected concepts: Study · Agent · Machine · EvidenceRead Runs and progress

Evidence

Sources, hand-ins, files, notebook records and review to inspect against the question. An output alone does not establish a validated finding.

Connected concepts: Study · Stages · Run · AristotleRead Evidence and results

Skills

Reusable instruction packs selected in an agent or Team’s capabilities. They do not supply credentials or connect a runtime.

Connected concepts: Workspace · Agent · TeamRead Skills

Connectors

Configured access to tools or data sources. Inspect availability and the agent’s capability configuration.

Connected concepts: Workspace · AgentRead Connectors

Aristotle

Helps you ask about selected workspace material and follow evidence. Its answers do not complete a study’s required review.

Connected concepts: Workspace · Study · EvidenceRead Ask Aristotle

Follow one piece of work

  1. Choose a workspace, the shared context for your lab's studies, colleagues and configuration.
  2. Create a study, a question or task with a description, assignee and stage scope.
  3. Assign an agent, a configured research role, or a Team, a coordinated group for a stage.
  4. Open Settings → Machines to confirm the runtime needed by your agent or Team is online. A machine is the computer exposing those runtimes; a run records one execution attempt.
  5. Inspect the resulting activity, hand-ins, files, notebook records and review.

A machine can expose more than one runtime. An agent's identity is reusable across work; its individual runs have their own progress and errors.

Use the scientific stages

The stage vocabulary is Grounding, Hypotheses, Experimentation and Validation. A study's configured scope determines which stages it covers. Stage labels describe the research process; they are not evidence that a particular result is correct. See scientific stages.

Add reusable support

Skills carry working instructions and supporting material. Connectors provide access to configured tools or data sources. Workflows package a specific experimental task. Aristotle helps you ask about workspace material and follow evidence. These serve different purposes; adding a skill does not connect a machine or supply database credentials.

Check your understanding

Open an existing study and identify its question, assignee, stage and most recent run. Then find a hand-in or a reported blocker. If you cannot identify the execution computer, use machines; if you cannot tell what was produced, use evidence and results.

Research, better together.