Execution layer for OpenClaw: multi-threaded AI workflows with multiple models and providers.
Website: https://parallelclaw.ai
MVP Blueprint: https://parallelclaw.ai/mvp-v2.html
Self-delegation skill for OpenClaw agents. The main agent decides on every query: direct reply or spawn a sub-agent.
- ⚡ Zero config — routing works immediately after install
- 🔥 True parallelism requires OpenClaw config check (see references/requirements.md)
- 💡 Fast Main + Smart Sub-agents — use fast model for main, powerful for sub-agents
# Via ClawHub (when available)
clawhub install talk-while-work
# Or download and install manually
wget https://github.com/parallelclaw/parallelclaw/releases/latest/download/talk-while-work.skill
clawhub install ./talk-while-work.skillgit clone https://github.com/parallelclaw/parallelclaw.git
cd parallelclaw
python3 skills/talk-while-work/scripts/install.py| # | Pattern | Status | Command |
|---|---|---|---|
| 1 | Talk While Work | ✅ Ready | — |
| 2 | Multi-Mind (/pc-ask) |
📝 Planned | Consensus voting across N models |
| 3 | Multi-Shard (/pc-scan) |
📝 Planned | Sharded throughput for mass tasks |
| 4 | Multi-Agent (/pc-do) |
📝 Planned | Stateful sub-agents with tools |
- Layer 1: Skills (zero infrastructure, open source)
- Layer 2: MCP-server (key vault, cost tracking, rate-limit pool, cache)
- Layer 3: Self-tuning routing (adaptive model selection, closed-source moat)
- Stage 1: OSS Wedge (free plugin)
- Stage 2: ParallelClaw Cloud ($10-50/mo, hosted orchestrator, BYOK)
- Stage 3: Personal Agent Layer (Memex integration, local-first memory)