Why This Exists
Cloud AI services create walled ecosystems — your memory resets per session, your data lives on their servers, and corporate policies decide what you can and cannot do. Switching providers means starting from scratch.
Salmex I/O takes a different approach: a local-first platform where you own your data, choose your LLM provider, and control what runs on your machine. It gets better the longer you use it, remembers across sessions and channels, and answers to you — not a corporate policy layer.
Architecture
Salmex I/O is built in Go with a SvelteKit frontend — a single binary with no npm supply chain risk. The platform is structured as eight composable layers, from channel ingestion through to the dashboard UI. Layer-by-layer detail: salmex.io/docs/architecture.
Channel Layer (Telegram, Slack, CLI, Web)Gateway (JSON-RPC, Auth, Routing)Agent (Reason, Act, Observe, Embedded Coder)Judge (Risk Tiers, Escalation, Audit Log)Memory (pgvector, BM25, Extraction, Decay)Plugin (JSON-RPC 2.0, Subprocess, Recovery)Scheduler (Cron, Natural Language, DLQ)UI (Dashboard, SSE, JSON-RPC API)
Platform Capabilities
Beyond chat, Salmex I/O is a full operations platform — it can search the web, write and execute code, browse websites, reach you on Telegram or Slack, run scheduled tasks, and remember everything across sessions and providers.
Persistent Long-Term MemoryMulti-Channel CommunicationEmbedded Coding AgentMulti-Engine Web SearchAES-256-GCM Encrypted ConfigNatural Language SchedulingBrowser AutomationExtensible Plugin System
Founder Responsibilities
As founder, I lead the venture end-to-end — product vision, technical architecture, full-stack implementation, go-to-market strategy, and community building. The platform is built independently, with no venture capital and no corporate roadmap.