Keep AI agents on current data: live database retrieval instead of stale copies

What Claix is

Claix sits between your agents and your database, and you set it up in three steps.

• Connect your database. Link it with a read-only role. Claix reads its structure and drafts a Data Map (tables, descriptions and relationships) for you to review and edit.
• Define the rules and retrieval. Create routes: each one lists the tables and columns it may use, plus a short guide with your rules and vocabulary. Then create an agent with a name, a business prompt that says when to use each route, and a response mode: raw results or reasoned answers in the type you ask for.
• Let your agent fly. Every agent gets its own endpoint, reachable over REST, MCP and A2A. You can run it on Claix Managed AI or bring your own LLM key.

The freshness problem

Most agent knowledge systems work well in the first demo. Then the business changes.

A price is updated. A booking is cancelled. A customer withdraws consent. An employee changes role. If the agent reads from a copy of that data, such as a vector index or an export, it keeps answering from the old version until the next sync. The answer sounds right and is wrong.

Copies also multiply: an old export, a newer one and the live table can all disagree, and the agent has no way to know which one is true.

How Claix keeps context current

• It reads live data. Each question runs against your database at the moment it's asked. There is no index to rebuild and no sync job to maintain.
• The Data Map follows your schema. Claix keeps a fingerprint of your database structure. When tables or columns change, the Data Map is flagged as outdated so you can re-scan and review only the differences, instead of starting over.
• Definitions are versioned. The Data Map and your routes are published as versions. Editing a draft never changes what a live agent uses until you publish it, and you can return to a previous version if something goes wrong.
• The endpoint stays stable. Your agent keeps the same URL, /retrieve/{agent_name}, while you change routes, rules and tables behind it. Nothing in your application needs rewiring.
• Every request is traced. You can see the question, the route that answered, the query that ran and the result, so you can check what an agent read and when.

Current data, and only the right data

Fresh data is not enough if an agent can read everything. In Claix, each route lists the exact tables and columns it can use, and an agent only sees the routes attached to it. Claix connects with a read-only role and validates queries before they run. Anything outside the routes you chose doesn't exist for the agent.

The business prompt adds guidance, not security: it tells the agent when to use each route and what your terms mean. What an agent can access is defined by its routes.

Example: a bookings assistant

A restaurant group connects its bookings database and creates an agent for its reservations assistant. The assistant sends a request:

[
  {
    "user_ask": [
      { "question": "Has the customer Javier Pérez accepted marketing consent?", "format": "boolean" },
      { "question": "What is Javier Pérez's next reservation: date, party size and table?", "format": "object" }
    ],
    "ia_response": [
      true,
      { "date": "2026-11-07T21:00:00", "party_size": 6, "table": "T-12" }
    ],
    "log_id": "9ff38974-1392-4c1a-83e5-3b2f09bb43dc"
  }
]

Later that day the customer withdraws consent and cancels the booking in the restaurant's own system. The next time the assistant asks, the answers are false and a different reservation, with no re-upload, no re-index and no change to the agent.

Integrations and cost

Each agent is available over REST, MCP and A2A, so it works as a tool from MCP clients, from agent frameworks and from automation platforms such as n8n. You can use Claix Managed AI, or connect your own LLM key to choose your model and lower the cost per question. Pricing is per question, with no plans to compare.

When Claix isn't the right fit

Claix works when your answers live in relational data and you can describe the relevant tables. It doesn't do semantic discovery over large piles of unstructured text; for that you need a search step in front. Access is scoped per agent, not per end user, so if each person must see only their own rows, enforce that in your database or application. And if your information is still in PDFs or spreadsheets, Claix Ingest can extract it into structured fields and load it into your database, so an agent can then ask questions about it.

Final takeaway

Agents shouldn't reason over yesterday's copy of your business. Keep your data where it is, tell Claix which parts each agent may read, and let every answer come from the current source, with a trace of what happened.