RunbookAI is an open-source CLI agent that investigates production incidents using hypothesis-driven reasoning. When an alert fires, it forms ranked hypotheses, runs targeted queries against your infrastructure (AWS, Kubernetes, CloudWatch), and systematically narrows to root cause.Key design decisions: - Hypothesis branching/pruning instead of broad exploration. Forms 3-5 hypotheses, tests...
https://github.com/manthan787/tapedeck
Hey HN, I built a terminal-based 4-track cassette recorder inspired by the Teenage Engineering OP-1. You can record from your mic, layer synthesizers, program drum beats, and mix everything down all without leaving the terminal.Architecture is three threads communicating via lock-free crossbeam channels: a 60fps ratatui UI thread, a control thread for app state, and a cpal audio callback at...
https://github.com/Runbook-Agent/RunbookAI
Hey HN, I built this project from personal pain while being on-call.You get paged at 3 a.m., open six dashboards, grep through logs, check recent deploys, and try to piece together what broke. Most of this is pattern matching that follows a predictable decision tree. Exactly what an AI agent should be doing.RunbookAI is an open-source project that can understand your infra, ingest runbooks and...
I built ContextLedger because I kept losing context between AI coding sessions. I'd wrap up a session with Claude or Codex, and the next day I'd spend the first 10 minutes re-explaining what I was working on, what failed, and what was left to do.
ContextLedger is a local-first CLI that captures events from your AI coding sessions (prompts, tool use, errors, files touched), summarizes them into structured "capsules," and generates resume packs / handoff documents so your next session starts with the right context already loaded.
Key design choices:
Local-first, SQLite-backed – everything stays on your machine. No cloud, no accounts.
Multi-agent – works with Claude Code (via hook events), Codex (rollout JSONL ingestion), and Gemini (coming soon!).
Privacy controls – prompt capture is off by default, secrets are redacted, and you can block prompt content from being sent to remote summarizers.
Summarizer-agnostic – supports local models via Ollama, or OpenAI/Anthropic APIs for generating session capsules and intent labels.
Handoff workflow – ctx-ledger handoff --agent claude --from latest generates context and can launch the agent directly with it injected.
It also ships with a web dashboard (ctx-ledger dashboard) that gives you a live view of your active agents and detailed analysis of your sessions — planning vs. execution time breakdowns, project-level time tracking, intent distribution, and tool usage stats. It auto-refreshes every few seconds and keeps syncing enabled integrations in the background, so it doubles as a monitoring panel while you work. The onboard command gets you set up in one step – it enables integrations, runs an initial sync, and starts the dashboard.
I've been using it daily for last few days and it's made a noticeable difference in how quickly I can pick up where I left off, especially on multi-day projects. The intent labeling and planning-vs-execution time breakdowns have also been interesting for understanding how I actually use these tools. Would love feedback.
https://github.com/openweavr/Openweavr
Hi HN!I’m building OpenWeavr, an open-source platform that lets you run AI agents on your own machines (Mac minis, laptops, or spare computers) to automate real-world tasks.The idea is simple: many workflows still require manual steps across apps, browsers, and tools. Instead of relying only on cloud SaaS automation, OpenWeavr lets you run agents locally so they can:• Automate workflows across...
Oh and I've been a long time Hacker News reader, but never posted comments. So I made a new account. There's nothing else to it.
Here's the link to the article: http://www.newsweek.com/assange-google-not-what-it-seems-279...