Kontext Cli

Kontext is the first tool we've seen that treats agent authorization like a networking firewall—policy checked at the call, not at the prompt. The local-first design is the right call: by evaluating every tool call on the machine, it keeps latency low and avoids the bottleneck of a gateway. The observe mode is a genuinely smart onboarding trick—you can backtest your policies against real traffic without breaking anything, then flip to enforce when you're confident. Where it really shines is teams running Claude Code, Codex, or Cowork on developer machines. If you've ever had an agent run rm -rf in the wrong directory or push to a protected branch, Kontext's deterministic blocks are exactly what you need. The audit trail is a compliance gift: every decision is attributed and exportable, so you can show who ran what, where, and why it was allowed. The Kestrel classifier is the headline grabber—22 microseconds per call, beating seven frontier LLMs on ShellRisk-Bench. That's a real, measurable advantage over any cloud-based classifier, where round-trip latency would slow agents down. But it's worth noting the benchmark is specific to shell commands; we'd like to see more diverse coverage. Kontext's biggest constraint is agent support. It only works with three agents. If your team uses a different tool, you're out of luck—no plugin SDK, no generic MCP hook. That's a hard limit that might push larger orgs to wait or look elsewhere. Pricing is reasonable for what you get, but the Pro tier at $149/mo might be steep for small teams. The free Starter plan covers developers on their own machines, so you can pilot it without spending. The Scale tier is aimed at higher volume and longer retention—if you need enterprise custom identity or deployment support, you'll have to talk