System Manual
Overview & Philosophy
What eClips Workforce is, how work is split between agents and people, and the mechanisms that make it safe to let agents run without someone watching each step.
What eClips Workforce is
eClips Workforce is an AI operations layer for a business. Specialized agents read the work that arrives every day, such as invoices, contracts, leads, tickets and job applications. They route each item by your rules, act through the tools you already use, and pass the exceptions to a person.
The unit of work is a run: one document or task handed to one agent. A run ends either in an action taken or in an exception queued for a person. The rest of this manual covers the parts that make that work: the architecture, the agent types, the harness, the trust model and the credit engine.
Who does what
Many automation tools wait for someone to click a button. Workforce agents work through whatever arrives, and people manage them the way they would manage a team: they set the rules and review the exceptions.
People set the rules, agents do the work
An agent reads, classifies, decides and acts through connected systems. A person writes the rules each agent follows, sets the budgets, decides how much each agent may do on its own, and reviews what lands in the triage inbox. Nobody has to approve routine work, and the policy behind it stays in human hands.
What makes it safe
Each of these has its own chapter in this manual.
- •
Every run is scored
Each run produces a confidence score. Anything below the auto-approve bar goes to a human triage queue. - •
Autonomy is set, and earned
Each agent type has a trust level per organization that an admin sets, starting at observe-only. Agents also build a track record: correct decisions relax approval gates, and errors bring them back. - •
Everything is metered and recorded
Every run is metered in credits and stored with its outcome, and every approval is stored with who gave it.
The confidence bar
Routing depends on the agent's confidence score. The thresholds are defined once, in CONFIDENCE_THRESHOLDS, and the orchestrator applies the same bar at every routing decision.
| 0.85 and above | The run can proceed and act through your connected tools, as long as nothing else flagged it. |
| 0.50 to 0.85 | The run goes to triage, with the reason that confidence was below the auto-approval threshold. |
| Below 0.50 | The run goes to triage, with the reason that confidence was below the minimum. |
Example · Three runs
A clean PDF invoice from a known vendor extracts at 0.93. That is above the bar, so the run carries on to the invoice processor. A smudged scan extracts at 0.71 and goes to triage with the proposed reading attached. A handwritten note with an unclear total scores 0.40 and goes to triage too, marked as below the minimum.
Who this manual is for
Anyone who operates, integrates with or evaluates eClips Workforce:
- Operators who configure agents, write rules and work the triage inbox.
- Administrators who manage plans, seats, member budgets and trust levels.
- Developers who integrate through the public API, or build on the MCP and Reach surfaces.
- Security and finance reviewers who need to check isolation, records and cost control.