Everything currently in flight
OpenClaw agent crew cockpit — task orchestration, model management, cost tracking, and ops monitoring.
Intelligent model routing — select models by capability, quota, cost, and task profile across all agents, prioritising free-tier providers to reduce DeepSeek spend.
Daily cost breakdowns, health summaries, memory trends, and anomaly alerts from Oscar — in canonical format.
Track server reboots and restarts — log boot events, uptime, and kernel history via journalctl.
Eliminate silent failures and stuck processes — make the crew self-healing with health monitoring and auto-recovery.
Three-layer test defense — Jest unit tests, Playwright E2E, and smoke tests for Mission Control stability.
Enable the crew to analyze images (screenshots, diagrams, mockups) and generate images on demand — cost-effectively, across multiple providers.
Snapshot the full agent ecosystem to GitHub — one bootstrap script to recreate everything on any machine.
Archived (7)
Design and maintain the optimal multi-agent crew — role definitions, dispatch rules, model assignments, and cost.
Wire DeepSeek's hosted API into OpenClaw as a provider so the crew can route work to cheap, capable models when local Ollama isn't enough but Anthropic is overkill.
Keep every project's documentation accurate and current so any agent or human can read the docs and instantly understand the system, instead of re-discovering it.
Structured knowledge base maintained by Kelly — bridges raw docs with query-time retrieval via cross-linked entries.
Run small open-weight LLMs locally on the home server's RTX 4060 (8GB) so cheap/routine agent tasks don't burn API credits.