> ## Documentation Index > Fetch the complete documentation index at: https://docs.usenash.com/llms.txt > Use this file to discover all available pages before exploring further. # Nash Agent > Nash Agent answers questions about your deliveries, investigates problems, and takes action — pausing for your approval on sensitive steps — all scoped to your organization and your permissions. Nash Agent answers questions about your deliveries, investigates problems across your orders and jobs, and takes action — pausing for your approval on sensitive steps. It can also analyze your operations to surface trends and answer higher-level questions. Nash Agent wears two hats. It's an **interactive assistant** — you ask, it answers and acts. And it's a **workflow-driven operator**: the same capabilities packaged as [custom agents](/nash-agent/custom-agents) that run on a schedule or when something happens, without anyone typing a prompt. Nash Agent is org-scoped and permission-aware: it only sees and acts on data within your organization, and only within the permissions of the person using it. ## What you can do Reach for Nash Agent when you'd otherwise be hunting through screens, exporting spreadsheets, or repeating the same checks every morning. A few concrete examples: * **Check a delivery.** "What's the status and ETA of order ABC-123?" or "Pull up the tracking for this customer's last delivery." * **Understand what went wrong.** "Why did this delivery go late?" — and get a timeline of what happened. * **Take an action.** "Cancel this delivery," "request a refund," or "reassign this to another provider." Sensitive actions pause for your approval first. * **Answer an analytics question.** "What was our on-time rate last week?" or "Compare provider completion rates this month" — answered from your own delivery data, with charts or a downloadable table. * **Put work on autopilot.** Stand up a [custom agent](/nash-agent/custom-agents) that runs a playbook on a schedule or when something happens, or an [improvement](/nash-agent/improvements) that pursues a goal over days. ## Where it lives Nash Agent lives in the Nash Portal, and you can also work with it in Slack and the Nash mobile app — so you can reach it wherever your team already operates. It can additionally reach out to customers and drivers by text or voice **on your behalf**, always with your confirmation. See [Channels](/nash-agent/channels) for how Slack, SMS, and voice work. ## How it works Every run — whether you typed a question or a schedule fired — follows the same loop: **observe, reason, act**. 1. **Observe.** The agent gathers context: your request, the live state of your operation, historical data, and anything your org has taught it through [knowledge](/nash-agent/knowledge) or [memory](/nash-agent/memory-and-learning). 2. **Reason.** It turns that context into conclusions — what happened, why it happened, what the options are — and recommendations with their assumptions stated. 3. **Act.** It closes the loop: answers your question, produces a report, proposes an action for your approval, or takes the action — and, if your org has enabled [memory and learning](/nash-agent/memory-and-learning), carries what it learned forward. A key part of the design is the **safety model**: read-only work (looking things up, running reports) happens automatically, while anything that changes your operation — cancelling a delivery, issuing a refund, reassigning a provider — pauses and shows you a confirmation first. By default, nothing destructive runs until you approve it; see [Guardrails](/nash-agent/guardrails) for how your org can tune the confirmation tiers. When a request is complex, Nash can break it into a visible plan, work through follow-up passes until it's done, and produce downloadable files (CSV, XLSX, PDF, or charts) from the results. What Nash can do for a given person depends on their role and permissions in your organization, and on your org's settings. If something isn't available, it's usually a permissions or settings choice — not a missing capability. ## Why we built it Running deliveries at scale means living in dashboards: clicking between screens to find one order, exporting data to answer a simple question, and repeating the same manual checks every day. The information is all there — it's just slow to get to, and slower to turn into a decision. Nash Agent is our answer: one conversational entry point for operational intelligence and action. Ask the way you'd ask a sharp teammate, and get back not just *what* is happening but *why it matters* and *what to do about it*. Ask for anything. When Nash can do the work, it does; when it can't, it says so plainly. ## Explore the docs Define reusable, scoped agents for the jobs your team repeats. Talk to a custom agent directly, under its own brief and permissions. Run agents on a schedule or in response to events, hands-free. Ground Nash with your SOPs, policies, and reference links. Set a multi-day goal and review recommendations as they come in. Let Nash carry context forward across conversations — opt-in and org-controlled. See what the agent is doing and what it costs, and cap spend. Reusable, named playbooks the agent pulls up when a situation calls for them. Which actions run automatically and which pause for your approval. ## Related How a fleet reports delivery state back to Nash. Build delivery operations directly against the Nash API.