Documentation
Glossary
The core concepts behind eClips Workforce. Terms link here from across the docs.
- eClips
- The company. eClips builds a family of products — eClips Workforce (the autonomic AI workforce for document operations) and more. "eClips" refers to the company/brand; each product carries its own name (e.g. eClips Workforce), so it's "eClips Workforce", never just "eClips", when you mean the product.
- eClipser
- Someone certified through eClips Academy on an eClips product. Complete a track — Workforce User, Administrator, or Developer — pass the exam, and you're a certified eClipser.
- Agent
- A specialized AI worker that performs one job — e.g. an invoice processor or contract reviewer. eClips Workforce ships 80+ agent types across departments (finance, HR, procurement, legal, ops, sales, IT/security, marketing).
- Agent run
- A single execution of an agent on one input. Every run is traced with a status, a confidence score, the model used, and the tokens/credits it consumed.
- Triage
- The exception queue. When an agent isn't confident enough or an action needs sign-off, the item is routed to the Approval Inbox for a human to approve or reject.
- Confidence score
- A 0–1 score an agent assigns to its own output. Bands: ≥ 0.85 proceed automatically, 0.65–0.85 proceed with a logged warning, below 0.50 always triage.
- Harness
- The runtime wrapper around every agent run. It handles retries with backoff, timeouts, the circuit breaker, model escalation/downshift, the evaluator on high-stakes agents, and the credit meter.
- Model downshift
- When a plan doesn't permit a premium model, the harness runs the agent on the best model the plan allows instead of blocking it — so every tier can use every agent, just on different brains.
- Model escalation
- When a low-confidence run warrants it (and the plan allows), the harness retries once on a more capable model (Claude Opus) for a better answer.
- Evaluator
- An independent grader that reviews the output of high-stakes agents (e.g. payroll, tax, fraud) and can force a triage or an Opus retry before the result is trusted.
- Trust level
- The degree of autonomy a workspace grants an agent (levels 0–4). Higher levels let more actions run without approval — but destructive actions are always gated.
- Context Book
- Your organization's institutional memory — vendors, approval hierarchy, SOPs, policies — that makes agents accurate. You set it up during onboarding and refine it over time.
- Credit
- The unit of usage. 1 credit = 1,000 tokens of a baseline (fast) model. Heavier models consume more credits per token. Each plan includes a monthly credit allowance.
- Credit multiplier
- The per-model factor applied to token usage when computing credits: fast (Gemini Flash, Claude Haiku) 1×, smart (Claude Sonnet) 12×, premium (Claude Opus) 60×.
- Member budget
- A per-user monthly credit cap that a workspace admin sets, so one person can't drain the shared pool. Available on Team, Growth, and Enterprise plans.
- Scope
- A permission attached to an API key (e.g. agents:run, triage:write, webhooks:read) that limits what the key is allowed to do.
- API key
- A secret token (prefixed eck_live_…) that authenticates REST API and SDK requests. The organization is derived from the key, so no organization id is sent.
- Reach
- eClips Reach — a browser-automation tool agents pick up when a target system has no API. It operates web UIs directly; every submit and destructive action requires human approval.
- MCP
- Model Context Protocol — the connector layer that lets agents call external tools and systems (ERP, CRM, accounting, ticketing, etc.) through a uniform interface.
- Organization
- A tenant in eClips. All data, credits, members, and settings belong to an organization, isolated from every other tenant by row-level security.
- Workspace
- Synonym for organization — your team's isolated environment, identified by a unique slug.