Open-source infrastructure
for AI systems
We build the tools that make autonomous agents actually work in production. Self-hosted, no vendor lock-in, MIT licensed.
AI agents are powerful in isolation
But in production, you don't run one agent — you run many. A coding assistant on your laptop, an autonomous agent on a server, automation workflows on another machine. They each maintain their own context and forget everything between sessions.
When one agent discovers something important, the others never learn about it.
We build the missing infrastructure layer.
[Memory]: Save pattern -> DB
Agent B (Server): Starting server sync...
[Memory]: Retrieve pattern <- DB
Success: Context unified across cluster.
Projects
Zengram
Shared memory for multi-agent AI systems
Your agents share a brain. Store a fact from Claude Code on your laptop, recall it from an autonomous agent on your server, get a briefing from n8n — all through the same memory system.
- Triple-path retrieval: vector similarity + BM25 keyword + entity graph traversal
- Typed memory with mutation semantics (events, facts, statuses, decisions)
- 15 MCP tools — works natively with Claude Code, Cursor, and any MCP client
- LLM consolidation engine — finds contradictions, merges duplicates, surfaces insights
- Entity relationship graph — automatic extraction and D3.js interactive visualization
- Client knowledge base — full client briefings via a single brain_client call
- Credential scrubbing — agents can't leak secrets into long-term memory
- 15 MCP tools — store, search, query, client, export, import, graph, briefing, reflect, and more
- Pluggable: OpenAI/Gemini/Ollama embeddings, SQLite/Postgres/Baserow storage
- MCP compatible — works natively with Claude Code, Cursor, and n8n
ZenVox
Voice to text. Cleaned by AI. Yours to keep.
Press a hotkey, talk, and cleaned text appears wherever your cursor is. Whisper runs locally — your audio never leaves your machine. Pick your own LLM for cleaning: Gemini free tier, Ollama for fully offline, or any of 5 providers.
- Faster-Whisper local STT — your audio never leaves your machine
- 5 LLM cleaning providers: Gemini, OpenAI, Anthropic, Groq, Ollama
- $0/month — Whisper is free, Gemini free tier handles cleaning
- VAD auto-stop, 4 cleaning presets, configurable hotkeys
- Bilingual FR/EN — native Franglais support
- CUDA GPU acceleration, searchable SQLite history
OpenClaw Memory Toolkit
Session logs to searchable long-term memory
Your OpenClaw agents generate thousands of lines of session data. This toolkit consolidates those logs into vector-searchable memory with automated deduplication, encrypted backups, and drop-in agent skills.
- Consolidation engine — LLM-powered fact extraction from raw session logs
- Documentation pipeline — ingest markdown docs into searchable vector memory
- 4 drop-in skills: memory-store, memory-query, memory-delete, docs-query
- Encrypted backups with GPG — automatic rotation, always recoverable
- Cron-ready — set it and forget it, runs on your schedule
Built in Production
ZenSystem isn't a research project. Every tool we release was built to solve a real problem inside a real business — then extracted, generalized, and open-sourced.
Zengram runs in production coordinating 10+ AI agents across 3 machines, powering client deliverables for a digital agency. ZenVox was built because $16/month for a Whisper wrapper is absurd. Both were born from the realization that the tools we needed didn't exist — so we built them.
About
Built by Steven Lefebvre. Founder of Expert Local, a Quebec-based digital agency specializing in local SEO and marketing automation.
ZenSystem is the technical arm — where internal tools become open-source projects.
Automate the repetitive. Protect the meaningful.