Frequently Asked Questions
General
What is Ariftly?
Ariftly is a Vertical AI Agent Platform. We ship autonomous agents that already understand their domain — AI compliance, sales outreach, and more — and let you deploy them in minutes rather than building workflows from scratch.
Agents that act. Not chat.
Is Ariftly only for technical teams?
No. Every Ariftly agent is built for the person who owns that domain. The AI Readiness Agent is for compliance leads, legal teams, and anyone answering procurement questionnaires. The Sales Agent is for sales reps and GTM teams. Future agents cover finance, operations, DevOps, and more. If you work in the domain, there's an agent for you.
How is Ariftly different from Zapier or Make?
Zapier and Make are workflow builders — you connect nodes and write the logic. Ariftly ships agents that already know what to do. You deploy them and extend their behavior by describing what you want in plain English. We do the implementation.
How is Ariftly different from LangChain?
Those are agent infrastructure tools — you build the agent yourself using their primitives. Ariftly is the finished product: agents with domain knowledge built in, human-in-the-loop approvals as a first-class feature, and a Skill system that lets non-engineers extend agent behavior without writing code.
Is Ariftly an AI assistant like ChatGPT?
No. Ariftly agents are autonomous — they initiate work, call tools, and take actions. They don't wait for you to prompt them. The human-in-the-loop approval system means they surface decisions for your review, but they drive the work forward on their own.
Agents
Which agents are available right now?
- AI Readiness Agent — EU AI Act compliance, procurement questionnaires, knowledge base audits. Live now.
- Sales Agent — Lead discovery, email enrichment, AI-drafted outreach. Early Access.
- Finance Agent, DevOps Agent, Founder Agent — Coming soon. Join the waitlist at app.ariftly.io.
Can I build my own agent?
Yes. The Remote Agent Protocol (RAP v1) is open and language-agnostic. Any HTTP service that implements the spec can be registered as an Ariftly agent. We provide a Node.js SDK to make this fast. → Build a Custom Agent
Do agents run continuously or on demand?
Both. Agents can be triggered manually, on a schedule, via API, via a Slack message, by events from other agents, or via MCP from Claude Desktop or ChatGPT. You configure triggers per agent.
What is BYOK?
Bring Your Own Key. You provide your AI provider API keys (OpenAI, Anthropic, etc.) in Settings → AI Keys. Ariftly routes your AI calls through Vercel AI Gateway using your keys. We never bill for AI usage — you pay your AI provider directly. Revenue is from platform subscriptions.
Can agents talk to each other?
Yes. An artifact produced by one agent can trigger a task in another agent via the event system. Cross-agent workflows are a first-class feature.
Skills
What is a Skill?
A Skill is a declarative extension to an agent's behavior, written in YAML. Skills attach to agent lifecycle hooks (before_task, after_task, on_artifact_emitted, etc.) and can emit artifacts, send notifications, auto-approve actions, or modify agent behavior.
Do I need to write YAML to create a Skill?
No. Go to Agents → [Your Agent] → Skills → New Skill, describe what you want in plain English, and the Skill Builder meta-agent generates the YAML. You review and approve it — no engineering required.
Approvals
What always requires approval?
Any action that affects the outside world:
- Sending an email
- Creating a pull request
- Posting a Slack message
- Publishing a compliance report to an external channel
- Adding contacts to a CRM
Can I configure auto-approval for certain actions?
Yes. Use Skills to define auto-approval conditions. For example, auto-approve follow-up emails to contacts who opened the first email, or auto-approve low-risk outreach within daily caps.
Where do approvals appear?
- Dashboard — the Approvals inbox shows all pending approvals across all agents
- Slack — if you configure a Slack approval channel, agents post the draft and you approve/deny with button clicks
Data & Security
Do you train on my data?
No. Your documents, questionnaire responses, and organizational data are never used to train any model.
Where is my data stored?
All tenant data is stored in isolated Neon Postgres databases (US East by default). EU data residency is available on Enterprise plans.
Do agents have access to my raw credentials?
No. API keys and SaaS OAuth tokens are stored in an AES-256-GCM encrypted credential vault. Agents receive short-lived proxy tokens scoped to specific actions — they never see your raw secrets.
Is the protocol open?
Yes. The Remote Agent Protocol (RAP v1) is a versioned, language-agnostic HTTP spec defined in JSON-Schema. It is designed to be public-safe from day one.
Pricing
How is pricing structured?
Per-agent subscription with daily task caps. See Pricing for current tiers.
| Plan | Price | Agents | Tasks/day |
|---|---|---|---|
| Free | $0 | 1 | 30 |
| Starter | $29/mo | 3 | 300 |
| Pro | $79/mo | All | 2,000 |
| Team | $199/mo + $49/seat | All | 20k/mo |
| Enterprise | Custom | Custom | Custom |
What counts as a task?
One execution of an agent action — responding to a questionnaire, discovering a batch of leads, generating a gap analysis report, or running a scheduled audit.
What does BYOK mean for billing?
If you bring your own AI keys (Starter and above), you pay your AI provider (OpenAI, Anthropic) directly. Ariftly charges only for the platform subscription — not per AI call and not per token.
Can I use multiple agents on one plan?
Pro and above include all available agents. Starter lets you pick up to 3 agents.
Is there a free plan?
Yes. The Free plan includes 1 agent and 30 tasks per day — enough to evaluate Ariftly with real data before committing to a paid plan.
Technical
What is RAP v1?
The Remote Agent Protocol v1 is the HTTP contract between Ariftly Core and every agent. It defines 6 endpoints every agent must implement. It's language-agnostic, versioned, and publicly documented. → RAP v1 spec
Does Ariftly support MCP?
Yes. Ariftly Core exposes agents and tools via the Model Context Protocol, so Claude Desktop, Cursor, and other MCP clients can invoke Ariftly agents directly.