Core concepts
The vocabulary Taizen uses, and what each object actually controls.
Taizen has a small number of objects. Learning these makes the rest of the product read easily.
Workspace and organization
Your organization is your company in Taizen: members, roles, credits, and the integrations connected at company level. Your workspace is where the data and the work live — calls, deals, documents, agents, bots, interviews.
Members have one of three roles — Admin, Editor, Viewer — described in invite your team.
Data sources and primitives
Everything Taizen reasons about reduces to four kinds of record:
| Primitive | Where it comes from | What uses it |
|---|---|---|
| Call | Gong, Fireflies, Genesys Cloud, roleplay sessions | agents, scoring, roleplay personas |
| CRM record | Salesforce, HubSpot — companies, contacts, deals | agents, interview targeting, bot building |
| Document | Uploads, document stores, Slack, Drive | agents, roleplay grounding, training plans |
| Interview | Win/loss voice interviews | battlecards, competitive briefs, coaching |
A data source is one connected origin of those records. Settings → Integrations → Data sources shows what is connected and when it last synced.
Agent
An agent is a saved unit of work with three parts:
- a trigger — a schedule, or an event upstream (triggers)
- a workflow — a name, a model, and the instructions it runs (build an agent)
- a delivery destination — Slack channels, email recipients, CRM write-back, or a Live App (delivery)
Agents are Active or Inactive. An inactive agent keeps its configuration and history but won't fire.
Run
One execution of an agent is a run. Every run stores its inputs, the steps it took, its output, its duration and its cost, and ends in a status — succeeded, failed, skipped, or succeeded-but-delivery-failed. Runs appear in the agent's Timeline, and the whole workspace's runs appear under Agents → Timeline.
A job is the scheduled unit behind a run: the thing that was due to fire. Agents → Jobs is where you look when a run you expected never appeared.
Skill
A skill is a reusable instruction document — markdown with frontmatter — that tells an agent how to do a recurring task: how your team writes a follow-up email, how you qualify a lead, how a battlecard is structured. Skills ship built-in and you can write your own. See skills and tools.
The distinction that matters: an agent's workflow prompt is what to do this run; a skill is standing knowledge it can pull in.
Bot, meeting, session
In Roleplay:
- a bot is an AI buyer — a persona built from a LinkedIn profile, a CRM contact, or generated as a synthetic buyer for an industry and job title
- a meeting is a prep setup for a specific upcoming conversation, with a stakeholder roster
- a session is one recorded voice roleplay, which can then be scored, replayed and shared
Scorecard, score, dimension
A scorecard (an evaluation template in the UI) is the rubric: dimensions, what counts as good, and how it's weighted. A score is the result of applying it to one call. A dimension is one axis within it — objection handling, discovery depth, value articulation.
The same scorecard applies to a practice roleplay and to a real customer call. That's deliberate: it's the only way to tell whether the drilling changed anything. See how scoring works.
Training plan, stage, pass gate
A training plan is a sequence of stages, each with steps, that escalates from a friendly buyer to a hostile one. A pass gate is the score a rep must reach on a stage before the next one unlocks, and a plan can end in a certification. See Training.
Document store
A document store is a named collection of documents that products read from — the grounding for roleplay buyers and training plans. Your playbook, your pricing, your competitive one-pagers. See document stores.
Live App
A Live App is an HTML page an agent publishes, at a stable URL, that keeps its data live — a dashboard, a report, a deck. A template freezes a design so later runs reuse it. See Apps.
Credit
Model work consumes credits. Settings → Usage shows consumption across the organization and the current billing cycle. Individual test runs in the agent builder show their own cost, so you can see what an agent will cost before you schedule it.