Glossary

Agent task: a pending step proposal created by an automation that requires your approval before executing. Appears in the task inbox.

Auto-execute: the auto_execute_safe_steps autonomy tier, where a side-effect-free step (like drafting a sequence) runs immediately without going through the task inbox. Contrast with draft only / approve each step.

Automation: a trigger-step pair that lets the agent automate pipeline tasks. Each automation has an autonomy tier controlling how much of a run happens without your approval, and an optional per-run cost cap.

Autonomy: the tier controlling how much of an automation's run happens without operator approval — draft_only, approve_each_step, or auto_execute_safe_steps.

Client: a company you are prospecting. The primary entity in the pipeline.

Client research (CR): one client's research within a run, containing findings, a report, a propensity score, and an outreach draft.

Cohort: a group of clients in a run segmented by propensity score range (high, medium, low). Visible in the rollup.

Company profile: a description of your own organisation used by the agent to assess prospect fit. Versioned, snapshotted onto each run.

Condition: the per-trigger-kind parameters (e.g. a score threshold, dwell days) that scope when an automation's trigger actually fires.

Confirmation: shorthand for the draft_only / approve_each_step autonomy tiers, where a proposed step goes to the task inbox and waits for human approval. Contrast with auto-execute.

Enrichment: the process of automatically populating a client record with public metadata: headcount, industry, LinkedIn URL, and logo.

Finding: a discrete, factual claim about a prospect extracted from public sources during a research run. Tagged to a topic and assigned a confidence score.

Methodology: a reusable scoring rubric defining topics, criteria per topic, and weights. Attached to an offering. Versioned.

Offering: the product or service you are selling. Each research run researches clients against exactly one offering. Versioned.

Outreach draft: an AI-written message anchored to the findings from a client's most recent research. Editable before use.

Propensity score: a 0-100 rating of how likely a client is to purchase your offering right now, derived by evaluating findings against the methodology rubric.

Research run: a batch job researching one or more clients against a single offering. Contains one client research record per client.

Rollup: a cross-client summary generated when all clients in a run reach a terminal state. Includes cohort distribution, top performers, and common themes.

Signal: an external event relevant to a tracked client, produced by a signal source (for example, a funding announcement from the company news source).

Signal source: an integration that polls an external data provider on a schedule and produces signals when relevant events occur.

Snooze: deferring an agent task to a later time. The task reappears in the inbox after the snooze period expires.

Step: what an automation does when its trigger fires and condition passes. The six step kinds are: draft_sequence, advance_stage, draft_outreach, re_research, notify_user, enrich.

Tenant: your workspace, a shared environment with its own clients, runs, offerings, and settings. Multiple users can belong to the same tenant.

Topic: a category in the methodology that groups related findings (for example, Technology fit, Budget signals, Buying team).

Trigger: the event that causes an automation to be evaluated. Manual (run-now only) plus five event kinds: cr_completed, score_crossed, stage_dwelled, signal_event, nth_day_no_touch.