FAQ

Questions about building, testing and deploying AI agents

Answers to the 36 most common questions about the AI agent lifecycle in Backprompter: building agents, validating them with simulated users and evaluator agents, deploying them, and charging for them.

Last updated November 18, 2026

4 questions

Getting started

What Backprompter is and who it is for.

What is Backprompter?

Backprompter is a platform for building, testing and deploying AI agents in one place. Domain experts and developers define an agent in plain language, test it with simulated users and evaluator agents, and promote the exact version that passed to production. Agents run in Backprompter's cloud or fully air-gapped on your own network. See the platform.

Who is Backprompter for?

Backprompter is for teams that need AI agents they can trust in front of customers or employees: developers and IT teams, domain experts such as support leads or HR specialists, enterprises that need company-wide rules, and agencies or independent developers who build agents for clients and want to charge for them.

Do I need to know how to code to build an AI agent?

No. You describe the agent's role, rules and tone in plain language, and built-in AI suggests prompts as you write. Developers can add tools, APIs, MCP servers and model settings, so experts who don't code and developers work on the same agent.

Does Backprompter build full apps or websites?

No. Backprompter builds, tests and deploys the AI agent. You build your app in Claude, ChatGPT or any development tool, then call the agent from it through the REST API, SDKs or MCP. If you don't need an app at all, you can publish the agent as a hosted chat, website embed or Slack app.

8 questions

Building AI agents

Creating agents, choosing models and connecting your data.

How do I build an AI agent in Backprompter?

You build an agent in four steps: describe it in plain language, click Test Now to chat with it, refine the prompt using AI suggestions and evaluation results, then Save and Commit a version. You promote the version that passes evaluation to production. How the prompt builder works.

Which LLMs can I use for my AI agent?

You can choose from the major models available through OpenRouter, set per agent, or bring your own OpenRouter keys. On your own network, agents can use local model servers such as LM Studio, Ollama or vLLM. Each agent in a team can run on a different model. Choosing models.

Can I build a multi-agent system?

Yes. You can combine specialist agents behind a router that sends each message to the right agent, and agent teams can call other agent teams as steps in a larger workflow. A long workflow that reaches its limit can continue from where it stopped. Simulated users can test the router and its specialist agents together.

Can my AI agent answer from my company's documents (RAG)?

Yes. Upload documents or URLs, and your agents answer from them and show their sources. Evaluations set each answer next to the matching passages from the knowledge base you provide, so the evaluator can check that responses are grounded.

What is BRAG?

BRAG (Backprompter RAG) gives your agents a connected memory of your documents. It finds how the documents in a collection relate, on its own, so agents give complete answers drawn from all of them and show where each part came from. Turn it on for files you have already uploaded with one click, see the connections on the memory map, and tick Deep search on an agent for the hardest questions. BRAG is included in the Teams and Enterprise plans. See pricing.

Can my agent connect to APIs, tools and MCP servers?

Yes. Define API tools and MCP servers once per app, then choose which ones each agent can use to look up live data and take action in your systems. Connection credentials are encrypted and never displayed after saving.

Can domain experts and developers work on the same agent?

Yes. App owners invite teammates as editors. Experts own what the agent says, developers own how it connects, and both work on the same agent and its versions in one place.

Is there help inside the builder?

Yes. The Backprompter Copilot is a panel next to the builder that answers questions about Backprompter and about the app you have open. It sees a summary of your screen (never your API keys), lights up the button or field to use next, can put a suggested prompt into a field, and reviews an agent's prompt on request. It never saves, deploys or starts paid work by itself.

10 questions

Testing and evaluating AI agents

Validating agent behavior before it reaches users.

How do I test an AI agent before deploying it?

In Backprompter you test where you build: chat with the agent using Test Now, run simulated users through everyday requests and edge cases, and score each conversation with an evaluator agent against your company rules and the agent's own rules. Then apply suggested fixes, retest and promote the version that passed. Test and evaluate.

What are simulated users in AI agent testing?

Simulated users are AI-driven test users that chat with your agent the way real people do, from routine questions to people who push back, change the subject or try to talk the agent out of its rules. In Backprompter, the simulated user and the evaluator are separate, so scoring stays objective.

What is an evaluator agent?

An evaluator agent is an AI agent that grades another agent's conversations. In Backprompter you build evaluators like any other agent, so you decide how they grade and what they return, from a single score to a detailed report.

What is a batch evaluation?

A batch evaluation runs many simulated conversations at once, across multiple mock users, agent versions and topics, and scores every one. It gives you an assessment of quality and latency, sets each answer next to the matching passages from your knowledge base so the evaluator can check grounding, and compares up to four models or versions side by side, without creating a version for each model by hand. Batch evaluation.

Can I evaluate an AI agent that runs somewhere else?

Yes. Connect an agent you host elsewhere, and Backprompter tests it with simulated users, scores it with your evaluator agents and runs batch evaluations on it, the same way as agents built in Backprompter. Test and evaluate.

How do I reduce AI agent hallucinations?

No platform can guarantee an LLM never makes things up, but you can catch most problems before users do. Ground the agent in your documents, define what it must always or never do in its rules, check grounding automatically with batch evaluations, and have evaluator agents test answers against your real policies before each release.

How do I find the lowest-cost LLM that still meets my quality bar?

Run the same tests on different models and compare the results against your own standards. In Backprompter, a batch evaluation compares up to four models or versions side by side, scored by the same evaluator, and you save the model you choose as a version, so you can keep frontier models for complex reasoning and move routers and FAQ agents to cheaper models. Test models, not just prompts.

How do I catch regressions when a prompt or model changes?

Re-test every change before it ships. In Backprompter every prompt change and model switch goes through the same simulated users and evaluators, every version is saved, and you can compare versions and roll back. Your app calls the agent, not the model, so a tested switch needs no app redeploy.

Can I enforce company-wide rules and guardrails on every AI agent?

Yes, on the Enterprise plan. Admins set company rules once, such as prompt-injection or personal-data-leak tests, and every agent built in Backprompter must pass them during testing, including agents built by teams outside IT. Teams can add rules but not remove them.

How does Backprompter suggest prompt fixes?

Backprompter suggests prompt changes based on the evaluator agent's findings and your own feedback. You can apply, discard or compare each suggestion line by line, then retest to confirm the fix worked.

4 questions

Deploying AI agents

Putting a tested agent in front of users.

How do I deploy an AI agent to production?

You promote the exact version that passed evaluation, then serve it wherever your users are: a hosted chat app with sign-in, a website embed, a Slack app, the browser SDK, the REST API or MCP. Every channel serves the same validated agent. Deploy and maintain.

Can Claude Code, Codex or other AI tools call my agent?

Yes. Backprompter can act as an MCP server, so Claude Code, Codex, Cursor, Claude Desktop and other MCP clients that can send an API key can call your agents directly, in addition to calls through the API and SDKs.

Do I need to build a backend to deploy an AI agent?

No. Hosting, end-user sign-in, per-user data, encrypted credentials and private-by-default data handling are part of the platform, so you can deploy a validated agent without building a server, database or login system.

Can I run AI agents air-gapped or self-hosted?

Yes, on the Enterprise plan. Backprompter runs on your own infrastructure with no internet connection required, using local model servers such as LM Studio, Ollama or vLLM, with offline licensing. Every prompt, document and user record stays inside your network.

2 questions

Maintaining agents

Keeping agents current after launch.

How do I update an AI agent after launch without breaking it?

Send every update through the same build and test loop as the first version. In Backprompter you re-test each change against your company rules and agent rules before you promote it, with versions, notes and rollback built in. Updates that don't break what works.

Can business teams update an agent without a developer?

Yes. The people who know the job, such as a support lead or benefits specialist, can change the agent, test it and promote it themselves, and the same standards protect every change. Agencies can hand an agent to a client's team the same way.

2 questions

Monetizing AI agents

Charging for the agents you build.

How do I charge users for my AI agent?

Connect your own Stripe account and charge end users directly. Sign-in and per-user data are already built in, so there is no payment system or user database to build. Payments go straight to your Stripe account, and Backprompter takes no share of your revenue (Stripe's standard processing fees apply). Earn revenue.

Can agencies and independent developers build agents for clients?

Yes. You can build an agent, use evaluation results to show the client it works, deploy it in the client's channels, hand it off to their team, who re-test every change, and charge end users through your own Stripe account.

3 questions

Security and data

How Backprompter handles your data.

Does Backprompter store my users' conversations?

No, not by default. Backprompter doesn't log conversations. If you want to keep conversations or other data, the SDK lets you save exactly what you choose.

Are my prompts and data used to train AI models?

In Backprompter's cloud, requests go only to model providers operating under zero-data-retention terms, so prompts and outputs are not retained or used to train their models. In an air-gapped deployment, nothing leaves your network.

How are API keys and credentials protected?

API keys, model keys and connection credentials are encrypted at rest and never displayed after saving.

3 questions

Comparing platforms

How Backprompter differs from other tools. Based on each vendor's public materials as of November 2026.

How is Backprompter different from Lovable?

Lovable is an AI app builder that generates full web applications from prompts. Backprompter focuses on the AI agent itself: how it behaves in conversation, how it is tested and how it is deployed. Lovable tests the app: it writes and runs tests for app logic as it makes changes, and can test user flows in a browser. Its documentation doesn't describe evaluating an agent's conversations with simulated users, evaluator agents and batch evaluation. Lovable's documentation also doesn't describe an air-gapped edition with local models. Many teams use both: build the app in a tool like Lovable, and call a tested Backprompter agent from it. Backprompter vs Lovable.

How is Backprompter different from Dify?

Dify is an open-source platform for building LLM apps and workflows, with cloud, self-hosted and enterprise editions. Backprompter's difference is that agent testing is built into the place you build: simulated users, evaluator agents you design, batch evaluation, suggested prompt fixes and company rules every agent must pass before release. Dify's documentation describes interactive testing, side-by-side model runs, built-in retrieval evaluations for knowledge bases, logs, and integrations with LLMOps tools, and a Dify blog post co-written with Arklex points to Arklex's open-source tool for simulated-user testing. Backprompter also includes sign-in for your app's users and Stripe billing, so you can charge the people who use your agent without building a payment system. Backprompter vs Dify.

Do I need a separate AI agent evaluation tool?

Not with Backprompter. Testing, evaluation, suggested fixes and deployment share one interface, so you don't export conversations to another tool, copy fixes back by hand or maintain another subscription and integration. Backprompter vs building it yourself.

3 questions

Pricing and plans

Plans, trials and what each includes.

How much does Backprompter cost?

There is a free trial, then three plans: Solo Dev at $5 a month, Teams at $129 a month with 25 people and 25,000 agent calls a month included, and Enterprise, priced on request. On top of the plan you pay for what you use from your prepaid balance: $0.015 for each agent reply and $0.01 for each lookup, such as a tool call, a web search or an answer grounded in your files. Each extra person is $29 a month. See pricing.

Is there a free trial?

Yes. The free trial lets you build, test and evaluate agents, including simulated users, evaluator agents and batch evaluations, with capped processing. Deploying a live agent requires a paid plan: Solo Dev, Teams or Enterprise. See pricing.

What does the Enterprise plan add?

Enterprise adds self-hosted and air-gapped deployment, the admin layer for company rules and guardrails every agent must pass, an auditor role that evaluates agents without changing them, 1,000 people and 250,000 agent calls a month included, unlimited agents and batch evaluations, single sign-on (SSO) and 24/7 support. Pricing is on request.

Try it on your own agent

Build and test agents free. Upgrade when your agent is ready to go live.