Sentō
Self-improving AI agents on your Claude subscription — no API bills
Most agent frameworks assume you have an Anthropic API key with a generous credit balance and a high tolerance for surprise billing. Sentō takes a different bet: what if your agents just ran on your existing Claude Pro or Max subscription instead? One command — `npx sentoagent init` — installs Claude Code, 17 plugins, and a full agent infrastructure stack. Agents launch in tmux, conduct an onboarding conversation to personalize their configuration, and then run 24/7 with persistent memory via ClawMem, web browsing via Playwright, and automatic self-healing restarts via Guardian Bot. The framework supports agent-to-agent communication, which is rarer than it should be at this layer. Platform support covers Discord, Telegram, Slack, and iMessage. Deployment targets include Linux VPS, Docker, macOS, and Windows via WSL. The "self-improving" claim refers to the agents' ability to adapt their memory and tooling over time based on usage patterns — not recursive self-modification in the sci-fi sense. At 2 GitHub stars this is extremely early, and the one-command-installs-everything approach should trigger some healthy skepticism before you run it on a production machine. But the core idea — that subscription-based agent infrastructure democratizes access without metered costs — is the right framing for where AI agents are heading in 2026.
Panel Reviews
“Running agents on Claude Pro instead of API credits is a genuinely different business model. No credit burn, no usage tracking, no surprise invoices at the end of the month. The tmux-based persistence is scrappy but it works, and the 17-plugin bundle means you get memory, browsing, and scheduling without assembling your own stack.”
“Two GitHub stars and a one-liner that installs 17 unvetted plugins on your machine. I don't care how clever the subscription arbitrage is — this thing is too new to trust with anything sensitive. 'Self-healing restarts' also means 'runs forever whether you want it to or not.' Hard pass until there's a meaningful community and an audit trail.”
“The subscription-over-API-credits model will become standard as Claude subscriptions get richer. Why pay per token when you're already paying flat-rate? Sento's architecture — persistent memory, self-healing, multi-platform — is what agent infrastructure looks like when it matures.”
“The pitch is compelling but the execution is too early. I want my 24/7 agents to have a proper dashboard, configurable guardrails, and more than 2 stars of community validation before I hand them my Discord and iMessage credentials.”