Skip to content
Back to work

LLM agent platform

Joshu

A CLI agent platform with MCP tool calling, semantic memory and an agent-to-agent API.

server over HTTP and SSE
A2Aserver over HTTP and SSE
non-regression on every change
CInon-regression on every change

The problem

Most agent demos stop at a single chat loop. Real use needs agents that call tools safely, remember across sessions, and plug into editors and other services.

What I built

  1. 1

    Orchestrated local and remote LLMs behind one CLI with secure tool calling over MCP servers.

  2. 2

    Added persistent semantic memory and multi-turn context management.

  3. 3

    Exposed agents through an agent-to-agent server (FastAPI, SSE) for IDE integration and external clients.

  4. 4

    Automated non-regression and security checks with GitHub Actions.

Outcome

  • A reusable agent base that other tools can call over HTTP, with CI guarding behavior on every change.

Have a similar problem?

Tell me about it. I reply within 48 hours with a first read of the problem and an honest opinion on fit.

Book a call