Skip to content

concepts.md

Latest commit

 

History

History
74 lines (53 loc) · 3.22 KB

concepts.md

File metadata and controls

74 lines (53 loc) · 3.22 KB

Mass: concepts

Mass is an AI-powered social simulation platform. You test ideas, concepts, products, pitches, and social experiments against many diverse AI personas from the command line. All usage is via the mass CLI; there is no API, GraphQL, or web UI.

Main concepts

Cohorts

A cohort is a named group defined by a text prompt (e.g. "UK parents, 25-45"). The LLM turns this prompt into a weighted cohort config: demographics, locations, and other dimensions. Cohorts do not contain persona text; they define who to generate. You then create one or more personas from that cohort (1-10 per run).

Personas

Personas are LLM-generated characters. Each has:

  • Name and a long backstory (the persona text used by the LLM when replying).
  • Structured metadata: age, gender, pronouns, location, job, education, personality traits, relationship status, and similar fields.
  • Optional extended metadata: lifestyle, health, values, tech savviness, communication style, and more.
  • Username (for mentions and identification).
  • Optional connections to other personas.

Personas are stored as one JSON file per persona under data/personas/<id>/<id>.json. See Data layout and Creating cohorts and personas.

Workspaces

Workspaces hold conversation state (messages). They are used by:

  • Chat: so you can continue a thread with a persona across invocations.
  • Report generation: the prompt and report context live in a workspace.

You can list, delete, rename, and fork workspaces (fork at a message to branch the conversation). See Workspaces and Chat.

Reports

Reports are batch runs that ask a cohort (or an explicit list of personas) a question and aggregate the results. Four types are available:

  • feedback: sentiment, quotes, summary, verdict (positive / neutral / negative).
  • debate: for/against views, persona positions, verdict.
  • questionnaire: extract questions from the prompt; each persona answers; aggregated view.
  • ideas: idea generation and aggregation.

Output is JSON (stored under reports/) plus optional HTML export. See Reports.

Flow

flowchart LR
  subgraph create [Creation]
    Prompt[Prompt]
    Cohort[Cohort config]
    Personas[N Personas]
    Prompt --> Cohort
    Cohort --> Personas
  end
  subgraph use [Usage]
    Chat[Chat]
    Report[Report]
  end
  Personas --> Chat
  Personas --> Report
Loading
  1. You provide a prompt (e.g. "UK parents, 25-45").
  2. Cohort create turns it into a cohort config and optionally creates personas.
  3. Persona create (if not done in one go) adds more personas to an existing cohort.
  4. You chat with a single persona or run a report over a cohort (or persona list) to test an idea or question.

Next steps