Mergen AI provides out-of-the-box AI agents that plug into the existing architecture of any company, SMB or enterprise. The offers are surfaced at the agent level, so customers can relate to what they are buying: a company that needs the ability to talk to its data never sees the underlying RAG architecture or harness of the Ask agent. From their point of view, they are buying the Ask agent — and paying per question-and-answer pair.
Ask agent
An agentic RAG implementation built on Postgres + pg_vector + AGE. Every answer is returned with three objects: Document, the original document that contains the answer; Fragment, the part of that document where the answer is; and Evidence, the proof that the answer really rests on the Fragment and that the Fragment really contains it.
Example
Pricing
$3 USD per run. Volume discounts: up to 10k runs, list price; 10–20k, 15–30%; 20–40k, 30–40%; 40–60k, 40–50%; 60–80k, 50%; beyond 80k, custom enterprise tier.
Research agent
A three-step research workflow built on a coding agent plus, when needed, the Ask agent. The default engine is Claude Code, but the workflow is vendor-agnostic. First, the model plans the research path given the topic, the depth and the available sources — web and/or internal documents. Second, it pulls in the information from those identified sources. Third, it generates a research report from that material in the available research template format.
Example
Pricing
$8 USD per run. Volume discounts: up to 5k runs, list price; 5–10k, 15–30%; 10–15k, 30–40%; 15–20k, 40–50%; 20–30k, 50%; beyond 30k, custom enterprise tier.
Work agent
This is where it gets interesting. The Work agent does the work that SMB or enterprise staff would do to get their functional objectives done: the multi-step actions behind a business objective, automated end to end. The output is a report, a ZIP dossier with multiple files, or both. The formats of the report, the dossier and the files inside it are predetermined and customizable, and the user iterates on the output until they are happy and sign off on the report page. The output can be consumed as is, or become the input to another workflow.
It is a multi-step workflow built on the neuro-symbolic AI loop harness. Depending on the chain, it may use the long-term memory service, the semantic context knowledge graph and the regulatory compliance service.
Example
Pricing
$21 USD per run. Volume discounts: up to 2k runs, list price; 2–4k, 15–30%; 4–6k, 30–40%; 6–8k, 40–50%; 8–10k, 50%; beyond 10k, custom enterprise tier.