🚧 STATUS: CURRENTLY UNDER DEVELOPMENT 🚧
Ground Control is a collection of open-source, lightweight CLI and TUI tools for OpenClaw enthusiasts who care about costs, efficiency, and running agents without lighting their money on fire.
These tools are built for people running OpenClaw on any resource-limited environment (and aren't afraid of the terminal). If you bought a MacMini, these probably aren't the droids you're looking for...
You know the YouTube tutorials where they say "I went ahead and bought a Mac Mini, but you can totally set this up on a VPS" and then proceed to demo a 200MB React app with Google Sheets as a backend... this project is the antidote to that.
Ground Control is mission control for the rest of us.
CLIs for agents. TUIs for humans.
AI agents don't need a pretty dashboard. They need fast, structured, low-token interfaces. A CLI tool reading from a SQLite database costs far fewer tokens per interaction. Having an agent talk to a Google Sheet? 2,000–4,000 tokens, easy. Multiply that across dozens of daily tasks, and you're hemorrhaging money on overhead, not on actual work.
Ground Control is built around a few core beliefs:
- Every token matters. If your tooling is burning tokens on bloated interfaces, you're paying a tax on every agent action. We minimize that tax.
- Lightweight by design. SQLite backends, Go binaries, no external API dependencies for cost data. No Docker compose files with twelve services. No Electron apps. No npm install that downloads half the internet.
- Terminal-native. The TUI isn't a compromise — it's the design philosophy. Built with Bubble Tea and Lip Gloss, these tools look good and run anywhere a terminal runs.
- Agents and humans share the same workspace. Agents use the CLI. Humans use the TUI. Both read and write to the same data. No translation layer, no API middleware, no nonsense.
- Zero network dependency for cost data. OpenClaw already logs every model response with full token counts. We ingest locally — no relay workers, no edge functions, no external API polling.
Ground Control is an umbrella project. The actual tools live in their own repos:
Cost Optimization Cost monitoring, spend tracking, and budget management in a beautiful TUI. Track daily/weekly/monthly spend across providers, break down costs by model, set budget alerts, and watch your burn rate in real time.
Telemetry ingests usage data directly from OpenClaw's session JSONL files — every model response from every provider (Anthropic, OpenAI, Google) is logged locally with full token counts. A lightweight ingestion service tails these files, computes estimated costs from a static pricing table, and writes per-request events to a local SQLite database. No external API polling. No relay infrastructure. Sub-millisecond reads.
Your agents monitor and analyze their own costs and evaluate them against OpenClaw events: new sessions, compacts, gateway resets, cache writes, etc. That's everything you need for a full agentic feedback loop to get your OpenClaw setup optimized for token-use. Your agents look at costs and events, find patterns, suggest improvements, and you approve/deny; all right in the TUI.
Features:
- Real-time cost tracking and analytics across Anthropic, OpenAI, and Google
- Per-request granularity — every model call attributed to the exact agent, session, and model
- Budget alerts with configurable thresholds
- Per-model and per-agent cost breakdowns and cache write ratio monitoring
- Daily, weekly, and monthly spend views with hourly burn rate
- Static pricing table (~90% accuracy, sufficient for optimization decisions)
- Approval workflow for agents to suggest improvements, and you to approve them
Task Management. Agent task management with Kanban and table views. Assign work to agents (or yourself), track status, set priorities, and manage due dates — all from the terminal. Your agents interact via CLI commands; you get the TUI overview.
Features:
- Kanban and table views with filtering
- Multi-agent support (assign tasks to specific bots or humans)
- Priority levels, due dates, and status tracking
- CLI interface for agent task creation and updates
- SQLite backend — no external dependencies
┌─────────────────────────────────────────────────────────┐
│ GROUND CONTROL │
│ │
│ ┌──────────────────────────────────────────────────┐ │
│ │ OPENCLAW GATEWAY │ │
│ │ │ │
│ │ Agents: Dex, Cassandra, Max, Borkus │ │
│ │ Providers: Anthropic, OpenAI, Google │ │
│ │ │ │
│ │ Every model response → session JSONL w/ usage │ │
│ │ ~/.openclaw/agents/*/sessions/*.jsonl │ │
│ └────────────────────┬─────────────────────────────┘ │
│ │ │
│ JSONL files │
│ (append-only) │
│ │ │
│ ┌────────────────────▼─────────────────────────────┐ │
│ │ INGESTION SERVICE │ │
│ │ (cron or file-watcher) │ │
│ │ │ │
│ │ Tail JSONL → Extract usage → Apply pricing → │ │
│ │ Write per-request events to SQLite │ │
│ └────────────────────┬─────────────────────────────┘ │
│ │ │
│ ▼ │
│ ┌──────────────┐ ┌──────────────┐ │
│ │ TELEMETRY │ │ DISPATCH │ │
│ │ (TUI/CLI) │ │ (TUI/CLI) │ │
│ └──────┬───────┘ └──────┬───────┘ │
│ │ │ │
│ ▼ ▼ │
│ ┌──────────┐ ┌──────────┐ │
│ │ SQLite │ │ SQLite │ │
│ └──────────┘ └──────────┘ │
│ │
│ ┌──────────────────────────────────────────────────┐ │
│ │ OPENCLAW AGENTS │ │
│ │ Interact via CLI · Low-token · Structured │ │
│ └──────────────────────────────────────────────────┘ │
│ │
│ All reads are local. Sub-millisecond. No network deps. │
└─────────────────────────────────────────────────────────┘
Agents use CLIs · Humans use TUIs · Data lives in SQLite
| Layer | Technology | Why |
|---|---|---|
| TUI Framework | Bubble Tea + Lip Gloss | Beautiful terminal UIs in Go. Fast, composable, no runtime dependencies. |
| Language | Go | Single binary deploys, tiny memory footprint, runs on anything. |
| Database | SQLite | Zero configuration, file-based, agent-friendly. No database server required. |
| Data Source | OpenClaw session JSONL | Usage data originates locally — no external API polling required. |
| Cost Estimation | Static pricing table | ~90% accurate, sufficient for optimization decisions. Optional reconciliation against provider APIs if needed. |
| Aesthetic | Late 1960s NASA Mission Control | Because constraints should become identity, not apologies. |
This is for you if:
- You are looking for a full cost-feedback loop for OpenClaw cost-optimization
- You're running OpenClaw on a VPS or resource-limited environment
- You care about token costs and want both human and agent-level visibility into your spend
- You prefer terminal tools over web dashboards
- You want your agents to use efficient, low-token interfaces
- You appreciate tools that do one thing well
This might not be for you if:
- You have an unlimited API budget and don't care about costs
- You want a drag-and-drop GUI
- You need enterprise-grade multi-tenant infrastructure
- You think Google Sheets is a database
Telemetry dashboard — Cost Monitoring in the Terminal
Each tool has its own installation instructions (for Humans) as well as "Agent Guides", and "SKILL.md" (if you're a bot) in its respective repo. But here's the quick overview:
# Install Telemetry (cost monitoring)
go install github.com/max-geller/openclaw-telemetry@latest
# Install Dispatch (task management)
go install github.com/max-geller/openclaw-dispatch@latestTelemetry's ingestion service needs access to your OpenClaw session directories (~/.openclaw/agents/*/sessions/). See the openclaw-telemetry README for setup details and pricing table configuration.
Pull up a chair. We saved you a seat.
Ground Control is a passion project born out of a simple frustration: OpenClaw can quickly become a runaway train of costs if you're not careful. What's worse, the OpenClaw ecosystem doesn't have enough lightweight, cost-conscious tooling for people running agents in resource-limited environments. We're looking to close the full cost feedback loop, with some guardrails.
- Bug reports and feature requests — If something's broken or missing, open an issue. No issue is too small.
- Code contributions — Whether it's a one-line fix or a whole new feature, PRs are welcome. Check the issues labeled
good first issuein each repo for a starting point. - New tool ideas — Have an idea for a tool that fits the Ground Control philosophy? Open a discussion. If it's lightweight, terminal-native, and agent-friendly, it probably belongs here.
- Documentation — Good docs make good tools. If you can explain something more clearly than we did, please do.
- Testing on different environments — Running on a Raspberry Pi? A different VPS provider? ARM? Let us know how it goes.
- Pricing table updates — Providers change model pricing 2-3 times per year. If you spot a pricing change before we do, PRs to update the seed data are always welcome.
- Safety First. Provide the right tools for your agents to course correct on costs. Nothing more.
- Keep it lightweight. If a feature requires pulling in a massive dependency or standing up a new service, it probably doesn't belong here. Every dependency is a token cost and a maintenance burden.
- Terminal first. If it can't run in a terminal, it doesn't ship.
- Think about tokens. Every interface an agent touches should be as token-efficient as possible. That's the whole point.
- Fork the relevant repo
- Create a feature branch (
git checkout -b feature/your-feature) - Make your changes
- Submit a PR with a clear description of what and why
- openclaw-telemetry — Cost monitoring TUI with local JSONL ingestion
- openclaw-relay — Task management TUI
- Agent configuration management
- Cross-tool integration (telemetry-informed dispatch decisions)
- Optional reconciliation layer (compare estimates against provider billing APIs)
- Web dashboard (SvelteKit, reading from VPS API or Litestream-replicated SQLite)
- Community plugin system
- Public documentation site
MIT. Use it, fork it, make it better.
If Ground Control helps you keep your agents under budget, consider giving us a ⭐. It helps other cost-conscious OpenClaw enthusiasts find these tools.