Kilo launches model-agnostic CLI 1.0 to embed AI across developer workflows

Kilo has reoriented its product strategy around a terminal-native command-line interface that moves AI capabilities out of single, IDE-bound sidebars and into any environment where engineers actually work. The new CLI is delivered as an MIT-licensed codebase and is explicitly engineered for sessions that span local editors, remote SSH environments, and incident-driven terminal work. A central technical feature is a persistent context layer that stores structured notes in Markdown inside repositories so agents retain state across sessions and interfaces, reducing the common problem of lost context. The client is deliberately model-agnostic: it can broker requests to a broad catalog of more than five hundred models, and it implements the Model Context Protocol to connect with external tools and internal services. Kilo ties this terminal experience to a separate Slack integration that can ingest multi-repository context and push fixes or pull requests directly from chat, aiming to reflect where engineering decisions frequently happen. On the economics side, Kilo sells usage credits at exact provider API rates and offers subscription tiers that mint bonus credits to improve the cost-per-output calculus for teams. Pricing examples include a $19/month starter tier, a $49/month professional tier, and a $199/month expert tier, each delivering additional credits; the company is also running a limited welcome bonus that increases initial credit balances. The product differentiates from polished but platform-bound offerings by leaning into portability and community contribution, arguing that open development speeds iteration and enables enterprise auditing. From a security posture, Kilo routes workloads through compliant cloud infrastructure options so proprietary code can remain within controlled perimeters while still connecting to external model services. Competitive levers are clear: Kilo trades the friction of vendor lock-in for a composable stack that lets teams pick models by cost and capability, whereas incumbent offerings emphasize integrated ecosystems and longer-lived agent sessions. The approach reduces a single-vendor dependency risk but introduces variability in model behavior and cost profiles that teams must manage. For organizations deciding between convenience and control, Kilo’s stack offers a pragmatic middle ground that amplifies automation while keeping architectural choices explicit. Adoption will hinge on how well the open CLI handles enterprise security requirements, integrates with existing CI/CD and monitoring systems, and whether the promised economics scale under heavy agent usage.