The Perplexity AI API is useful when an application needs search, researched answers, or access to models. For customer support, the key question is not simply whether an answer includes citations. It is whether the answer uses the right information for that customer and your business.
As checked on September 8, 2026, Perplexity recommends its Agent API for new projects. Its Sonar migration guide says Sonar Chat Completions is supported until September 27, 2026. If you are following an older tutorial, check the endpoint, model names, and response parsing before copying its code.
This guide covers the API choices, the responsibilities that remain with your application, and a separate option for support operations: using eesel CLI to configure and work with an existing teammate.
What is the Perplexity AI API?
Perplexity’s current quickstart describes four core APIs. They serve different purposes, so “send a question to Perplexity” is not a complete implementation choice.
| API | Main purpose | What your application receives or uses |
|---|---|---|
| Agent API | Web-grounded answers and agent workflows | Generated output, sources, and tool activity |
| Search API | Search without an LLM-generated answer | Ranked web results |
| Router API | Model access through a shared endpoint | Output from the selected model |
| Embeddings API | Semantic search and related text tasks | Vector representations of text |
A research assistant might use Agent API to produce an answer with sources. A custom retrieval pipeline might use Search API and handle synthesis itself. Embeddings serve a different job again: they represent text for similarity search, rather than answer the user’s question.
What changes from Sonar to Agent API?
The older Sonar Chat Completions pattern sends a messages array and reads choices. Agent API uses input and a typed output array. The result can contain different item types, including the answer and search results.
That means migration is more than selecting another model name. Review how you send context, request search, and extract sources. Perplexity’s migration documentation also describes presets and tools, including web search, URL fetching, MCP, and custom functions.
Perplexity is not limited to public-web answers. MCP and custom functions can extend an Agent API application with private data and business tools. That access still needs authorization and an implementation. It does not automatically connect your helpdesk or establish which internal document should take precedence.
Getting started without exposing credentials
Generate an API key in the Perplexity API Console, choose the API for your task, and follow its current quickstart. Use an environment variable or secret manager rather than embedding the key in frontend code or committing it to a repository.
Begin with a small set of representative requests. For research, check whether the cited pages actually support the answer and whether their dates fit the question. For an action-taking application, test what happens when the required information or permission is missing.
Perplexity also offers a terminal CLI. Its pplx command returns structured search results for shell workflows and coding agents. That is useful for gathering external information, but it is a different tool from eesel CLI, which operates an eesel teammate.
Pricing and reliability: Measure your own workload
Use Perplexity’s pricing documentation for the chosen API and model. Costs can include model tokens and tool or request charges, depending on the route you use. A short user question does not necessarily mean a short execution when an agent performs several research steps.
Pricing differs across APIs and models. Inspect actual usage and compare ordinary requests with long, tool-heavy ones before forecasting volume.
For reliability, separate three checks: did the request complete, did the answer follow the source, and was that source appropriate? A citation makes verification possible; it is not proof that the answer is correct.
Your application also needs timeout handling, rate-limit handling, retries, and monitoring. If tools can change business records, ensure a retry cannot silently repeat an action. Those are application responsibilities, not reasons to dismiss a research API as unsuitable for production.
Where company support knowledge changes the decision
Consider a customer asking whether an opened item can be returned. A general web result about consumer returns may be useful background, but it is not necessarily your policy. The answer may depend on the item, purchase date, region, and exceptions approved by your team.
You can build those connections and rules into a custom application. Or, if the desired outcome is a support teammate working with existing tools, evaluate a helpdesk teammate that already provides that product experience.
The distinction is responsibility, not “public data versus private data.” Perplexity can participate in a private-data application. eesel supplies the support-focused teammate; its CLI gives developers and coding agents a way to configure and inspect that teammate.
Use eesel CLI for company-grounded support work
eesel CLI lets you inspect connections, upload knowledge, review instructions, chat with a teammate, and examine its activity from a terminal. It operates the same agent and workspace as the dashboard, not a separate terminal-only assistant.
That makes it useful alongside a developer’s existing research workflow. A coding agent can gather public documentation through a research tool, then use eesel CLI to inspect the support setup. Decide explicitly which information belongs in company knowledge; do not automatically turn web research into policy.
With Node.js 18.17 or newer, start by logging in to your existing workspace:
npx @eesel/cli login
npx @eesel/cli whoami
npx @eesel/cli agents
Choose the intended agent and replace AGENT_ID below with its ID:
npx @eesel/cli integrations --agent AGENT_ID
npx @eesel/cli status --agent AGENT_ID
npx @eesel/cli instructions --agent AGENT_ID
These commands show the connected systems, whether connected content has downloaded, and the teammate’s standing rules. They do not establish that every source is current policy or that every answer will be correct.
Add an approved policy and check its use
For the returns example, upload the document your team has approved:
npx @eesel/cli files upload ./approved-returns-policy.pdf --agent AGENT_ID
The upload adds knowledge. It does not change standing instructions or resolve contradictions with another source. Review the instructions separately, including when the teammate should ask for more information or escalate to a person.
Then test representative questions on an agent with appropriate action restrictions. Include one question the policy answers, one missing a required detail, and one asking for an exception. Inspect the answer and activity. CLI chat can take actions, so treat it as interaction with the teammate rather than an inherently read-only search.
For helpdesk knowledge, inspect the relevant integration setup. eesel supports connections such as Zendesk and Freshdesk; some setup steps require browser authorization.
Let a coding agent inspect results
The CLI prints JSON, so Claude Code, Cursor, or Codex can read a command’s result and decide what to inspect next. Give it a bounded task such as “check the selected agent’s sources and instructions, report conflicts, and ask before making changes.”
That instruction limits the coding agent’s task. It is distinct from eesel’s approval process for actions awaiting review. The CLI also exposes activity, automations, and approvals so you can inspect ongoing work and review those actions.
Use --help for command options. For writes, --dry-run prints the server call without sending it; it does not simulate an answer. Headless scripts can use EESEL_API_URL and EESEL_API_TOKEN, with explicit agent scope. Keep these credentials separate from your Perplexity API key and out of source control.
This is not a built-in Perplexity-to-eesel integration. You decide whether the tools belong in the same workflow and review any information transferred between them.
Choose based on the work you want to own
Choose Perplexity’s APIs when you need research, search, model access, or a custom agent application. Measure its performance on the sources and tasks your users need.
Choose eesel when you want to operate a support teammate with connected company knowledge. Its CLI lets technical users configure and inspect that product while the support team uses the same dashboard.
Try eesel. Once your workspace and teammate are set up, use the CLI to inspect the knowledge and instructions for one support scenario before expanding the rollout. Check current eesel pricing separately; using the CLI does not move its usage onto your Perplexity account.
