agent-memory-server
PublicSelf-hosted, git-backed memory any AI agent can plug into
TypeScript
0
2026 - now
mutheejj / README.md
Hi, I'm John. I build AI agents that do real work.
Full-stack engineer in Nairobi. Laravel, TypeScript and Python, with agents that call real tools and tests that prove they recover.
- Contributions, last 12 months
- 10,789
- Public repositories
- 62
- Current streak
- 11 days
- Writing published
- 17
AI agents that do real work
Tool-calling agents wired into real systems, with memory, human approval for anything irreversible, and evals before they ship.
Agent infrastructure
Memory servers, parallel worktree orchestration, chaos tests and scheduled jobs: the plumbing that makes agents reliable.
Full-stack platforms
HR, hiring and marketplace platforms in Laravel, Next.js and React Native, with tests, CI and deployment from day one.
African fintech integrations
M-Pesa Daraja payments, callbacks and reconciliation, exposed to agents through MCP servers.
Pinned
All 14 projectsworktree-orchestrator
PublicParallel coding agents in isolated git worktrees, merged deterministically
TypeScript
0
2026 - now
mcp-african-markets
PublicMCP servers for African fintech: M-Pesa payments, reconciliation, sandbox by default
TypeScript
0
2026 - now
agent-test-harness
PublicChaos testing for AI agents: inject failing tools, grade the recovery
TypeScript
0
2026 - now
Andor
ProductAn AI coding agent that lives inside VS Code
TypeScript
1
1
2024 - now
AjiraWorks
Client workA work and gig-economy platform connecting talent to opportunity
PHP
2024 - now
10,789 contributions in the last year
Busiest day: 489 on Sep 29
Oct
Nov
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Languages across public repos
- Python 25%
- TypeScript 17.3%
- Java 13.5%
- C 11.5%
- JavaScript 9.6%
- Shell 5.8%
Recently pushed
This site speaks MCP
My agents draft articles and newsletter issues here through a Model Context Protocol server. I review, then publish.
$ claude mcp add --transport http portfolio \
https://john.muthee.mutexai.net/mcp \
--header "Authorization: Bearer jm_..."
# then, in a session
> draft this week's build notes from my commits
create_post(type: "newsletter", status: "draft")
✓ Draft saved for reviewStack
- TypeScript
- PHP
- Python
- SQL
- Dart
- Java
- C
- React
- Next.js
- React Native
- Tailwind CSS
- Alpine.js
- Blade
- Laravel
- Node.js
- REST APIs
- Queues
- WebSockets
- PostgreSQL
- MySQL
- Redis
- SQLite
- Vector search
- Claude API
- Tool use
- MCP
- Agent orchestration
- Evals
- RAG
- Linux
- Docker
- Nginx
- GitHub Actions
- Cloudflare
Latest writing
All writing
Article
6 min read
Git-backed memory for AI agents: why files beat a vector database
I run several coding agents that share one durable memory made of markdown, JSON and git commits. Here is why plain files beat a vector store, and how sessions, leases and decisions work.
Note
1 min read
Why the assistant's tools default to read-only
A short note on a naming habit from the mutexai.net assistant. Read-only tools get plain verbs. Anything that writes gets a verb that says so, on purpose, so a reviewer never has to guess.
Article
5 min read
Running AI coding agents in parallel with git worktrees
Three agents on one checkout means trampled files and lost work. Giving each task its own git worktree, its own model and a deterministic merge step fixed that for me.
Article
4 min read
Chaos testing AI agents: inject failing tools, grade the recovery
Happy-path demos tell you nothing about what an agent does when a tool returns a 500. I test agents like distributed systems: inject faults on purpose and score recovery against bluffing.
Article
4 min read
repo-doctor: diagnosing failing test suites with an agent
Clone, install, run the tests, then separate real regressions from environmental noise. Why I built repo-doctor as a deterministic pipeline first and an agent second.
Build notes
A short email about building agents and the systems around them. No hype, no tracking pixels.
Latest: Build notes #4: this portfolio is live, and I can no longer write my own posts