Skip to main content

​
Frequently Asked Questions

No. Keys stay in the local WDK wallet and in the RGB Lightning Node you point the tools at. Trades settle as atomic HTLC swaps on Lightning, so the maker is a counterparty to the trade and never a holder of your funds.What changes with AI is not custody, it is who initiates a spend. That is what the confirmation gates and risk limits exist for.
The mnemonic is passed to the server process as an environment variable (WDK_SEED) or read from agent.config.json. Signing happens inside that process. The model calls tools and receives their results, so the seed is not part of the prompt or the tool output.That is not the same as the seed being safe. Anything with read access to the config file or the process environment has your mnemonic, which is why a mainnet seed does not belong in a file you commit or share.
It depends on which tools you call.
  • No node needed for market data (l402_*) or the Spark wallet (spark_*), which only needs a seed.
  • A node is required for the RGB Lightning Node tools (wdk_*) and for atomic swaps, since the taker side of the HTLC runs on your node.
The KaleidoCLI can provision a Docker-backed node, and kaleido-mcp exposes kaleido_node_* tools that drive that lifecycle for you.
It varies by surface:
  • KaleidoAgent reasons with a hosted model, Claude or OpenAI, selected with AGENT_PROVIDER.
  • KaleidoMind runs the model on the device through the QVAC SDK, locally or on a desktop you explicitly paired. Its tiered funnel means most requests never reach the model at all.
  • MCP servers are model-agnostic. Whatever model your MCP host runs is the one calling the tools.
The tools can reach mainnet, but treat it as the last step rather than the first. Start on Regtest or Signet with a throwaway seed, keep dry_run enabled on KaleidoAgent until the dry-run decisions look correct, and only then switch networks.Note that the different surfaces carry different maturity: KaleidoAgent and the MCP servers drive live funds directly, and an LLM making a wrong tool call is a real failure mode.
Three separate cost lines:
  • LLM tokens for hosted models. KaleidoAgent reports token cost per run in GET /status.
  • Zero inference cost for KaleidoMind, since the model runs on your own hardware.
  • Bitcoin fees in every case: routing fees on Lightning, on-chain fees when a channel opens or closes, and the maker’s spread on a swap.
Several independent controls, and it is worth using more than one:On KaleidoMind the equivalent is structural rather than configured: fund-moving tools are marked requiresConfirmation and pause for the host’s confirmation sheet.
Yes. That is what the Machine Payments Protocol tools cover. The agent probes a gated URL, receives an HTTP 402 challenge with a Lightning invoice, pays it, and submits the credential to get the data plus a receipt. No signup and no API key. The flow is on the MCP Servers page.
An MCP server provides the tools: typed, callable operations against a wallet, the DEX, or a node.A skill is a SKILL.md playbook that scopes which of those tools the model may reach and carries the execution plan. Changing a skill changes behaviour without touching runtime code. One is capability, the other is policy plus plan.
Yes. Every repository is public under github.com/kaleidoswap: the MCP servers, KaleidoAgent, KaleidoMind, and the skills themselves. See Additional Resources for the direct links.
The Desktop App hosts KaleidoMind directly, as in-app chat and as a namespaced MCP and CLI surface, switchable from Settings > Capabilities. The Browser Extension ships its own built-in conversational agent for wallet operations. The standalone MCP servers are for external hosts such as Claude Desktop or your own client.

​
Get Help

Check the Troubleshooting for errors rather than questions. For anything else, report a problem through your preferred channel from the options below, including:
  1. Which surface and version (MCP server name, KaleidoAgent commit, or Desktop App version)
  2. The network you are on (Regtest, Signet, or mainnet)
  3. The tool name that failed and the error text
  4. Host and runtime versions (MCP client, Node.js)
  5. Your config with the seed and any keys removed

Telegram Community

Ask the community.

GitHub Issues

Report a bug on the relevant repository.

Email Support

Direct support for urgent issues.