Infrastructure for AI Agents

Open-source tools to build, store, and operate AI agent fleets. From the database layer to the operations platform.

What We Build

Available Now · v1.1.0

PondDB — The Database for AI Agents

A lightweight, open-source database built for AI agent memory. Five memory types, cross-agent sharing, full audit logging, and MCP server built in. Works with any framework — OpenAI, LangGraph, CrewAI, Google ADK, Claude.

  • 5 memory types (working, episodic, semantic, procedural, shared)
  • Cross-agent memory grants with access control
  • MCP server for tool-use integration
  • 1,280+ tests, benchmarked (903 writes/sec)
  • SQLite-powered, self-hosted, Docker ready
$ pip install ponddb $ ponddb serve --port 8080 PondDB running on :8080 MCP server enabled 5 memory types ready
View on GitHub →
Coming Soon · In Development

PondOps — DevOps for AI Agents

Build, test, deploy, and optimize your AI agent fleet. Connection holding, auto-suspend, cost attribution, A/B testing, and CI/CD for agent builders. The missing operations layer.

  • Connection holding (hold requests while agents resume)
  • Auto-suspend / resume (save costs on idle agents)
  • Per-agent cost attribution and token budgets
  • Prompt A/B testing with canary rollout
  • CI/CD pipeline for agent configs
$ pondops proxy start --fleet ./agents/ Proxy listening on :8080 3 agents registered (2 suspended, 1 active) Connection holding enabled
Star on GitHub →

How It All Fits Together

Your Application
        │
        ▼
PondOps Proxy ← connection holding, routing, cost tracking
        │
        ▼
Agent Fleet ← suspend/resume, A/B testing, CI/CD
        │
        ▼
PondDB ← agent memory, grants, audit logs

PondDB is the storage layer. PondOps is the operations layer. Together, they give your agent fleet a complete infrastructure stack.

Open Source, Always

Both PondDB and PondOps are open source under the MIT license. Built by developers, for developers.

Get in Touch

Have questions, want to collaborate, or interested in contributing? Reach out.

tianlu@cedarpond.ai