The building blocks for AI agents, in your database

Semantic cache, agent memory, data integration, context gateway, intent router, agent playbooks, Kivi Lens tracing, and Agent Mesh support: eight building blocks for AI agents that keep their data...

Eight building blocks for the parts every agent needs: answers it has given before, what it remembers, the data and tools it may use, where a message should go, the answers you approved, and a trace of each turn. Each keeps its data as keys in your own KiviDB database. Seven are services in KiviDB Cloud; the eighth, Agent Mesh, is Redis's open-source library, tested on KiviDB.

One page for each

  • Semantic cache: Serve a model's answer again for a prompt that means the same thing.
  • Agent memory: Conversation history and lasting memories, found by meaning.
  • Data integration: Keep PostgreSQL and MySQL tables live in KiviDB.
  • Context gateway: One MCP endpoint over your records and your own MCP servers.
  • Intent router: Send each message to the right agent, by meaning, with no model call.
  • Agent playbooks: Approved answers and skills that agents discover by meaning.
  • Kivi Lens: OpenTelemetry traces of every turn, kept in your database.
  • Agent Mesh: Redis's open-source library for agent handoffs, tested on KiviDB.

One agent turn, start to finish

  • Pick the agent The intent router matches the incoming message to the closest intent and returns its target: the agent, workflow, tool or URL you set.
  • Give it data and tools The context gateway hands that agent one MCP endpoint: search tools over your records, and your own MCP servers behind it, limited by its key.
  • Give it what it knows Agent memory recalls the user's history and lasting memories; agent playbooks give it the answers and skills a person approved.
  • Answer repeats from the cache The semantic cache serves a stored answer when a new prompt means the same as an earlier one, before a model is called.
  • Trace the whole turn Kivi Lens records the turn as a trace, with KiviDB's own cache, memory, context, intent and playbook calls inside it.
  • Hand work to another agent Agent Mesh, Redis's open-source library, passes tasks between agents over streams in the same database, and lets a client resume a stream from its cursor.

Your database, your cloud

  • Your database Each keeps its data as keys in a KiviDB database of yours, on the cloud and in the region you chose for it.
  • One console Each service has its own section in the KiviDB Cloud console, next to your databases.
  • From an assistant The KiviDB MCP server has tools for the semantic cache, agent memory, the intent router, agent playbooks and Kivi Lens, and the context gateway is an MCP endpoint itself.
  • Start free A free 100 MB database needs no card.

Questions

What does it cost?

No extra charge beyond the database for the semantic cache, agent memory, the context gateway, the intent router, agent playbooks, Kivi Lens and Agent Mesh: their data uses the database's own memory. Data integration is $0.04 per pipeline-hour, plus $0.10 per GB written past 10 GB a month. Hosted models, if you choose them, are billed per token.

Where does the data live?

In your own KiviDB database, as keys you can see. Embeddings and model calls go to the provider you choose: your own key, our hosted models, or, for the intent router and agent playbooks, a model that runs inside KiviDB.

Do they work on sharded databases?

The semantic cache does. Agent memory, the context gateway, the intent router, agent playbooks and Kivi Lens run on databases without shards for now. Agent Mesh has not been tested on a sharded database.

Which KiviDB version do I need?

New databases run KiviDB 1.2.0. Agent Mesh needs 1.1.3 or later, for XADD ... IDMP. Each page lists anything else a building block needs.

Build your agent on KiviDB

Create a free database, then add a building block to it from the console.