Chibi bridges the gap between powerful but complex tools like
kubectland heavy visual dashboards like Lens. It lives in your terminal, understands your cluster via natural language, and predicts failures before they happen.
Kubernetes troubleshooting is traditionally painful. Engineers bounce between kubectl get events, kubectl logs, and Prometheus dashboards, manually correlating data to find the root cause of a crashing pod.
Chibi solves this by acting as a Principal SRE in your terminal. It automatically fetches cluster state, correlates it with live Prometheus telemetry, and feeds it into an LLM (OpenAI, Anthropic, Ollama) to give you an instant, actionable diagnosis.
- AI-Powered Diagnostics: Ask "Why is the payments pod crashing?" and get a deterministic, context-aware answer.
- Predictive SRE Daemon: Chibi runs in the background, analyzing event streams via a local SQLite state database to predict failures before they trigger PagerDuty.
- Zero-Trust WASM Plugins: Extend Chibi securely. Third-party plugins execute in a strict WebAssembly (
wazero) sandbox with zero host filesystem/network access. - Live Terminal Dashboards: Beautiful, interactive TUI built with Bubble Tea, featuring live CPU/Memory sparklines natively in your terminal.
- Autonomous Remediation: Configure strict RBAC policies allowing Chibi to automatically heal known issues (e.g., auto-restarting degraded deployments).
- Telemetry Correlator: Competitors just feed raw YAML to LLMs. Chibi actively queries Prometheus, merging live time-series data with static YAML for profound context.
- WASM Sandboxing: CLI plugins are a massive security risk. Chibi's
wazerointegration means you can run community plugins with mathematical certainty they aren't stealing yourKUBECONFIG. - True Terminal Native: We didn't build a web app. Chibi boots in <200ms and renders 60fps interactive dashboards directly in your favorite terminal emulator.
go install github.com/chibi/chibi/cmd/chibi@latestDownload the latest release from the Releases page.
- Configure your AI Provider:
export OPENAI_API_KEY="sk-..."
- Launch Chibi:
chibi
- Diagnose a Pod:
Inside the TUI, type:
Chibi will fetch the pod manifest, correlate CPU metrics, and stream the AI's diagnosis.
diagnose pod frontend-deployment-6f9b8c
chibi- Launches the interactive Bubble Tea TUI.chibi diagnose <resource> <name>- One-shot AI diagnosis.chibi watch- Starts the predictive daemon in the foreground.chibi plugins list- Lists installed WASM plugins.chibi config- Manages provider credentials and sandbox policies.
Chibi is built in modular Go:
- CLI Layer: Cobra & Viper.
- TUI Engine: Bubble Tea, Lip Gloss, Bubbles (Sparklines).
- K8s Layer:
client-gowithristrettocaching. - AI Layer: Provider router (OpenAI default) + Telemetry Correlator.
- Plugin Runtime:
wazeroWASM sandbox. - State Engine: Pure-Go SQLite (
modernc.org/sqlite) for event anomalies.
Configuration is stored in ~/.chibi.yaml.
ai:
provider: openai
model: gpt-4o
observability:
prometheus_url: http://prometheus-operated.monitoring:9090
remediation:
auto_apply_deployments: trueChibi natively supports:
- OpenAI (
GPT-4o) - Recommended for deep logic. - Anthropic (
Claude 3.5 Sonnet) - Excellent for YAML analysis. - Ollama - For highly secure, local, air-gapped clusters.
Security is paramount.
- RBAC: Chibi uses your local
kubeconfigand inherits your exact RBAC permissions. It cannot bypass cluster security. - Zero-Trust Plugins: External plugins cannot access the network.
- Read-Only Default: Chibi acts in read-only mode by default. Autonomous remediation must be explicitly enabled per resource type.
| Feature | Chibi | kubectl | Lens |
|---|---|---|---|
| Speed | <200ms | <100ms | >2000ms |
| Interface | TUI | CLI | GUI |
| AI Native | Yes | No | Paid |
| Predictive | Yes | No | No |
| Plugins | WASM (Secure) | Binary (Unsafe) | Node (Heavy) |
We welcome contributions! Please read our CONTRIBUTING.md and SECURITY.md before submitting PRs.
- Version 1: Reactive Diagnostics & Terminal UI
- Version 2: Predictive Daemon, Prometheus Integration, WASM Sandbox
(Note: V1 and V2 are fully implemented and stable).
Apache 2.0. See LICENSE.
Built for the Modern Kubernetes Era (2026)
Research & Product Vision: Aaryan Rawat