Agents
Conduit ships with a library of pre-built agents — named, opinionated Claude configurations that encode a specific workflow. Agents appear in Claude Desktop’s prompt picker once your organization is connected. This page explains the concept.
What an agent is
Section titled “What an agent is”An agent is a named prompt plus a structured output contract plus a curated set of vendor tool calls. When your technician runs an agent, Claude already knows:
- Which tools to call. The agent names the vendor integrations relevant to the workflow and the order to query them in.
- How to interpret the data. Each agent carries vendor-specific domain knowledge — for example, the ticket-dispatcher agent knows that Autotask Priority 4 means Critical and Priority 1 means Low.
- What to produce. Every agent declares a specific output format (a dispatch report, a QBR value-conversation pack, a security scorecard) rather than leaving the shape of the answer to the model.
The difference between “ask Claude about backups” and running the
backup-health-monitor agent is the difference between a freestyle
session and a production report.
When to use an agent vs. a plain chat
Section titled “When to use an agent vs. a plain chat”Use an agent when:
- The workflow is repeated regularly (daily triage, weekly reporting, monthly audits).
- The output has a known structure and consumers downstream (a client, a PSA ticket, an audit package).
- Multiple connected vendors must be queried in a specific sequence to get a coherent picture.
Stick with a plain chat when:
- The question is exploratory (“what’s going on with this one ticket?”).
- The tools involved are a subset of any one agent’s scope.
- You need latitude to pivot mid-conversation.
Agent categories
Section titled “Agent categories”Conduit’s agent library is organized by the vendor category the agent primarily serves, plus a portfolio category for agents that span multiple vendors.
| Category | Purpose | Example agents |
|---|---|---|
| RMM | Fleet health, patch compliance, backup audits | rmm-health-auditor, backup-health-monitor, patch-compliance-reporter |
| PSA | Ticket dispatch, SLA tracking, contract renewal | ticket-dispatcher, contract-renewal-tracker, service-desk-ops |
| Documentation | Audit completeness, detect stale runbooks, find orphaned assets | documentation-auditor, runbook-freshness-auditor, asset-documentation-linker |
| Security | SOC triage, endpoint hardening, threat correlation | soc-alert-investigator, threat-correlation-analyst, endpoint-hardening-auditor |
| Email security | Threat investigation, policy audits, user risk reports | email-threat-investigator, vap-reporter, tenant-policy-auditor |
| Incident management | On-call audits, incident command support | incident-commander, on-call-scheduler, post-mortem-writer |
| CRM / Sales | Pipeline review, quote assembly, margin analysis | pipeline-health-reporter, quote-builder, margin-analyzer |
| Marketplace | License optimization, renewal calendar | license-optimizer, renewal-calendar, license-manager |
| Accounting | Billing reconciliation, cash flow analysis | billing-reconciler, cash-flow-analyzer, profitability-reporter |
| Productivity | Identity audits, license optimization | identity-auditor, license-auditor |
| Portfolio | Span multiple vendors; highest-value workflows | client-360-briefer, qbr-prep-agent, renewal-risk-analyzer, security-posture-scorer |
Portfolio agents
Section titled “Portfolio agents”Portfolio agents are the differentiator — they produce outputs no single-vendor agent can because they cross-reference three or more connected systems. Typical portfolio workflows include:
- Pre-call briefs — aggregate PSA, RMM, security, and CRM context for a single customer in one query.
- QBR preparation — prepare a value conversation for the business owner (ease of doing business, emotional value, transformational value), with PSA, RMM, and security numbers kept in an appendix.
- Renewal risk — score every customer’s churn risk from SLA performance, engagement, commercial signals, and unresolved technical debt.
- Security posture — a vCISO-style composite score across identity, endpoint, email, training, and documentation signals.
- Incident war room — a live situation report that pulls from every relevant system the moment a P1 is declared.
If you run portfolio agents frequently, connect more vendors. Their output quality is directly proportional to how many relevant systems are wired up.
How agents resolve permissions
Section titled “How agents resolve permissions”Agents run in the same permission context as the user triggering them. An agent cannot see vendors the user does not have access to, and every tool call the agent makes is audit-logged under the user’s identity. If an agent requests a tool the user’s role allowlist forbids, Claude sees the denial and adapts its plan.
See the permissions reference for the full role matrix and tool allowlist model.
Writing your own agent
Section titled “Writing your own agent”Your organization can extend the agent library with custom agents that encode workflows specific to your operations.
Related
Section titled “Related”- Permissions — how agent authorization resolves.
- API reference — every agent is also addressable by prompt ID through the MCP prompt listing endpoint.