What Open Agent is
Open Agent is the Kiebot AI customer support layer for product teams that already pay for LLM access and do not want to pay a per-conversation markup on top of that. You bring your own OpenAI, Anthropic, or Google key, drop a snippet on your site, and the agent is live in about ten minutes. The bundled assistant handles the easy traffic; your humans pick up the hard questions from a single inbox with built-in CRM and real-time visitor alerts.
Three problems Open Agent set out to solve
- The double-billing trap. Most customer-support SaaS charges per conversation while also charging you for the LLM behind it. Teams that already have model access end up paying twice.
- Long-running conversations. Real support tasks span minutes, hours, or days. The runtime must survive restarts, retries, and human pauses without losing context.
- Trust on the operator side. Support managers need to understand what the agent did and why. Every decision must be auditable, and humans must be able to take over cleanly mid-thread.
The five components inside Open Agent
- Planner. Turns a customer message into a sequence of steps. Re-plans when reality disagrees with the plan.
- Memory. Short-term scratchpad for the active conversation, plus long-term episodic memory across sessions backed by a vector store.
- Tool catalogue. Tools (knowledge-base lookup, ticket creation, refund draft, etc.) described with structured schemas. Retrieved on demand so the model only sees what it needs.
- Executor. Runs tool calls with retries, rate limits, and per-step cost limits.
- Safety layer. Input filters, output filters, PII redaction, and human approval for irreversible actions like refunds or cancellations.
The BYOK design choice
Open Agent runs on your own LLM key by default. You pay for software (predictable monthly), not for tokens (variable per customer message). Teams already paying for OpenAI or Anthropic credits see the per-conversation cost drop sharply.
What customers ship on Open Agent
- Pre-sale chat that handles pricing, demos, and qualification before a human is paged.
- Post-sale support that resolves tickets with citations from your help centre.
- In-product onboarding assistants that nudge users through activation steps.
How to start: Pick one well-scoped support flow. Three tools. Add a human approval step before anything destructive (refunds, cancellations, plan changes). Ship that, measure, then expand the tool catalogue.
Where Open Agent sits in the broader ecosystem
The self-hosted AI agent space in 2026 has substantial public OSS in it: OpenClaw, Hermes Agent by Nous Research, plus Letta, Cheshire Cat, Anything LLM, and a long tail of others. Those are excellent for personal or developer-led use cases. Open Agent is built for the narrower job of customer support, with the operations and CRM features that on-page customer support actually needs.
Related reading
For the lower-level orchestration pattern, see LangGraph for Stateful Agents. For the broader stack, see The LLM Stack in 2026. For the public self-hosted agent landscape, see three one-click paths to a self-hosted AI agent.
