Question 1

What is AI Workforce?

Accepted Answer

AI Workforce is an enterprise AI agent platform that lets you deploy autonomous AI agents to handle customer interactions, operational workflows, and business processes. Unlike chatbots or simple automations, agents can reason, make decisions, execute multi-step workflows, and collaborate with each other — all while keeping humans in the loop when needed.

Question 2

How do I get started?

Accepted Answer

Sign up for a free account, create your first agent, configure its personality and knowledge, connect your integrations (email, Freshdesk, Slack, etc.), and define what tasks it should handle. You can deploy your first working agent in under 30 minutes.

Question 3

Do I need technical skills to use the platform?

Accepted Answer

No coding is required for most configurations. Agents, tasks, workflows, and integrations are all configured through a visual interface. For advanced use cases (API webhooks, database queries, custom integrations), basic technical knowledge is helpful.

Question 4

What's the difference between an agent and a task?

Accepted Answer

An agent is the AI entity with a personality and set of capabilities. Tasks define what the agent can do — each task has its own prompt, tools, workflow, and output actions. One agent can have multiple tasks (e.g., a support agent with tasks for billing, shipping, and general inquiries). Messages are routed to the right task automatically.

Question 5

What is the publish/draft system?

Accepted Answer

All configuration changes (tasks, workflows, actions) are saved as drafts until you explicitly publish them. Publishing creates an immutable snapshot (revision) of the agent's configuration. You can roll back to any previous revision instantly. The agent in production always uses the published version.

Question 6

What channels does AI Workforce support?

Accepted Answer

Email (SendGrid, Amazon SES), Freshdesk, WhatsApp, Slack, SMS, web contact forms, and custom API webhooks. All channels flow into a unified inbox where agents and human operators can see the full conversation history.

Question 7

How does the workflow engine work?

Accepted Answer

Workflows are multi-step processes that you design visually. Each step has three phases: Entry (execute an action), Process (AI reasoning), and Exit (decide the next step). Steps can branch based on AI decisions, wait for external events (like a customer reply), loop with safety limits, and escalate on timeout.

Question 8

What is the knowledge base used for?

Accepted Answer

The knowledge base stores information that agents can query during every interaction — product catalogs, support procedures, company policies, technical documentation. You can add content manually, upload files (PDF, DOCX), or import entire websites via sitemap discovery. Content is automatically chunked and embedded for semantic search.

Question 9

How do funnels and pipelines work?

Accepted Answer

Funnels are customizable Kanban boards that track contacts through any process — sales pipeline, onboarding flow, support escalation path. Agents can move contacts between stages automatically as part of their workflows. Every stage transition is logged with timestamp, actor, and metadata.

Question 10

What is the ontology engine?

Accepted Answer

The ontology engine lets you define domain-specific entities (orders, products, customers) with typed properties and relationships. When agents process messages, they can extract structured data from unstructured text and store it as a queryable knowledge graph.

Question 11

Can agents work together?

Accepted Answer

Yes. Agents can delegate conversations to other agents. For example, a front-line support agent can hand off a billing question to a specialized billing agent, which has its own tools, knowledge, and workflows. The conversation context is preserved across the handoff.

Question 12

What are scheduled actions?

Accepted Answer

Scheduled actions run on cron expressions (e.g., every hour, weekday mornings, monthly). They can execute any integration action — query a database, check for shipping errors, send reports. When the action returns a list, fan-out mode processes each item independently as a separate event.

Question 13

Which AI models does AI Workforce support?

Accepted Answer

We support OpenAI (GPT-4o, GPT-4 Turbo, GPT-3.5) and Google Gemini (2.0 Flash, 1.5 Pro). You can configure different models for different tasks — use a faster model for routine classification and a more capable model for complex reasoning.

Question 14

Can I use my own API keys?

Accepted Answer

Yes. You connect your own OpenAI or Gemini API keys. You only pay for the tokens you actually use, and your data is subject to your own API agreement with the provider — not ours.

Question 15

How does AI Workforce use AI?

Accepted Answer

AI is used in several ways: processing messages and composing replies, classifying and routing messages to the right task, making workflow branching decisions, extracting structured data from text, embedding content for semantic search, and supervising agent output via Guardian Agents.

Question 16

Is my data used to train AI models?

Accepted Answer

No. Your data is never used to train any AI model. When you use your own API keys, your data is subject to the provider's API data policy (both OpenAI and Google state that API data is not used for training). AI Workforce itself does not store or use your data for any purpose other than executing your configured agents.

Question 17

How does the platform handle AI hallucinations?

Accepted Answer

Several layers of protection: knowledge base grounding, structured output with JSON schema validation, Guardian agents that supervise outputs, escalation for uncertain responses, and simulation mode for safe end-to-end testing.

Question 18

What integrations are available out of the box?

Accepted Answer

AI/LLM (OpenAI, Google Gemini), Email (SendGrid, Amazon SES), Messaging (WhatsApp, Slack), Helpdesk (Freshdesk), CRM (HubSpot, Freshsales), Shipping (Shipmatic), Database (PostgreSQL, MySQL), and custom REST API webhooks.

Question 19

Can I connect custom systems?

Accepted Answer

Yes. The API integration accepts webhooks from any external system and can trigger agent workflows. You define which event types and actions are allowed. For outbound calls, agents can use the API action to send requests to any REST endpoint.

Question 20

How does the Freshdesk integration work?

Accepted Answer

Freshdesk sends ticket webhooks to AI Workforce. The agent receives the full ticket data (requester, subject, description, conversation history, attachments), processes it through its task pipeline, and can reply, forward, create notes, or close tickets — all via the Freshdesk API.

Question 21

Can agents send and receive emails directly?

Accepted Answer

Yes. SendGrid inbound parse and Amazon SES/SNS handle incoming emails. Agents can send outbound emails through either provider. Full email threading is supported (Message-ID, In-Reply-To, References headers).

Question 22

How do I add a new integration?

Accepted Answer

Go to Settings > Integrations, click 'Add Integration', choose the type, and provide the required credentials. Most integrations can be configured in under 5 minutes.

Question 23

Where is my data stored?

Accepted Answer

Data is stored in a PostgreSQL database in your deployment environment. File attachments use Active Storage (configurable: local disk, S3, or any cloud storage provider). You choose where your data lives.

Question 24

Is AI Workforce self-hosted?

Accepted Answer

We offer both cloud-hosted and self-hosted options. The self-hosted version runs on your infrastructure with full control over data, network, and access. Available on the Enterprise plan.

Question 25

How does the Guardian Agent work?

Accepted Answer

Guardian agents are supervisory AI instances that review planned output actions before they execute. They check for policy compliance, tone consistency, data accuracy, and risk level. If the guardian flags an issue, the action is blocked and an escalation is created for human review.

Question 26

What audit trail is available?

Accepted Answer

Every event, action, LLM call, and workflow step is logged with full parameters, results, timing, and cost. Logs are retained for 7 days (Free), 90 days (Pro), or custom retention (Enterprise).

Question 27

Can I control what agents are allowed to do?

Accepted Answer

Yes, at multiple levels: each task defines which output actions are available, simulation mode prevents real execution, guardian agents supervise outputs, the publish/draft system prevents accidental changes, and escalation triggers on any failure or uncertainty.

Question 28

Is there a free plan?

Accepted Answer

Yes. The free plan includes 1 agent, 3 tasks, 500 LLM calls/month, 5 workflow executions/day, and basic integrations. No credit card required.

Question 29

Can I upgrade or downgrade at any time?

Accepted Answer

Yes. Plan changes take effect immediately. When upgrading, you get instant access to the new features. When downgrading, the change applies at the end of your current billing period.

Question 30

What happens if I exceed my LLM call limit?

Accepted Answer

LLM calls are paused until the next billing cycle. Your agents stop processing new events, but existing data and configurations are preserved. You can upgrade anytime to resume.

Question 31

Do you offer annual billing?

Accepted Answer

Yes. Annual billing saves 15% on the Pro plan (EUR 69/month instead of EUR 79/month).

Question 32

Is there a free trial for Pro?

Accepted Answer

Yes. 14-day free trial with all Pro features, no credit card required. At the end of the trial, you can upgrade to Pro or continue on the Free plan.

Question 33

What counts as an LLM call?

Accepted Answer

Each call to the AI model counts as one LLM call — this includes message processing, task routing, workflow decisions, entity extraction, and guardian supervision. Embedding calls are counted separately.

Question 34

What tech stack is AI Workforce built on?

Accepted Answer

Ruby on Rails 8 with PostgreSQL, Sidekiq for background jobs, Tailwind CSS for the interface, and Stimulus JS for interactivity. The platform runs on standard infrastructure — no specialized hardware required.

Question 35

Is there an API for programmatic access?

Accepted Answer

Yes. The API integration accepts REST webhooks for event ingestion and action execution. Authentication uses API tokens. You can trigger workflows, query agent status, and receive results programmatically.

Question 36

How do workflow timeouts work?

Accepted Answer

Each workflow step can define a timeout duration and a transition target. If the step is in a waiting state and the timeout elapses, the workflow automatically transitions to the configured target — which can be another step, an escalation, or the end of the workflow.

Question 37

Can workflows loop?

Accepted Answer

Yes. Workflow transitions can point back to previous steps, creating loops. Each step has a configurable max_loops limit to prevent infinite execution. When the limit is reached, the step's escalation behavior determines what happens.

Question 38

How does contact deduplication work?

Accepted Answer

Contacts are identified by email address or phone number. When a message arrives from a known email/phone, it's matched to the existing contact. The platform also supports bulk contact merging.

Question 39

What happens when an agent fails?

Accepted Answer

Failures trigger the escalation system automatically. The escalation record includes what failed, the input context, any planned actions, and the error message. If async retry is available, the system retries up to 2 times with exponential backoff. Persistent failures are escalated at critical risk level.

Frequently Asked Questions

Getting Started

Platform & Features

AI & Models

Integrations

Security & Data

Pricing & Plans

Technical

Still have questions?