[01]
Sub-agent dispatch
Spin up an isolated sub-agent for explore, plan, code, or verify. Each runs its own tool loop and reports back.
> ringpilot_
An AI coding agent. Ours.
Plenty of these exist. This is the one we built for Ringmaster's codebases — Java, PL/SQL, Angular, Python — with the tool loop and flows we needed.
PS1PowerShell · Windows 10 / 11
iex (iwr 'https://api.ringpilot.it/install.ps1' -UseBasicParsing).Content
In beta. macOS & Linux installers coming soon.
// what it does
Not a chatbot. A coding agent.
Real tool loop, sub-agent dispatch, memory across sessions. The basics of a coding agent — done in one binary, sized for our codebase.
[02]
15 tools, one loop
File edits, shell, search, dispatch — verified at every step. The agent acts on your code, doesn't just suggest.
[03]
Memory that survives
In-session compaction with retention scoring. Long-term facts keyed by files and tags. The agent remembers what it learned across sessions.
[04]
Self-updating binary
Thin launcher + Nuitka-compiled agent. CI tags a release, every dev's on latest by morning. No pip install dance.
[05]
OpenAI, for now
Routed through our LiteLLM proxy with quota tracking and centralized logging. More providers as we wire them up.
[06]
Quota in the status bar
Live token + spend tracking. Budget warnings before you hit the ceiling.
// flows
Coding agent. And an orchestration engine.
Same loop, same tools, same proxy. What CrewAI, LangGraph, AutoGen do as separate libraries, RingPilot runs natively — YAML missions that map sub-agents across your codebase, verify their work, and resume from the last good phase when something breaks.
mission: assess_dependency_upgrade parameters: [package_name] phases: - task: find_usage goal: | Find every file that imports {package_name}. output: JSON list of file paths tools: [bash, file_write] - task: assess_per_file over: find_usage goal: | Read {item} and assess whether the upgrade breaks it. output: JSON: { file, breaks, reason } verify: | Every "breaks: true" entry must cite a specific API change. - task: triage_report inputs: assess_per_file goal: | Write a per-team migration plan to {output_file}. tools: [file_read, file_write]
/flow run <mission> [param=value …]
// in action
A real flow, end-to-end.
No demo magic. This is what happens when you run a flow: parameter in, parallel scan, classification, triage report on disk. Hover to pause.