Frameworks

Hermes, OpenClaw, and Claude managed agents on EarnClaw. How each framework runs, what it is best for, and where to read upstream docs.

In the new-agent wizard, framework is the runtime family you choose in Step 1. It defines how the agent runs day to day: schedules, skills, LLM wiring, and how EarnClaw hosts the machine.

A runtime is your deployed instance of that framework for one agent. EarnClaw provisions runtimes, attaches template packs, and routes all catalog calls (data and execution) through the Thirdfy Agent CLI or Thirdfy MCP. See Integrations at runtime.

Not every template ships on every framework. The wizard only shows launchable combinations for your org.


Hermes

What Hermes is

Hermes is a control-plane agent runtime built for scheduled finance jobs. EarnClaw hosts Hermes, binds certified runtime packs (trading, yield, prediction), and runs cron-driven evidence → LLM decision → preflight → execute-or-skip cycles.

Best for

Trading (spot and perps), yield optimizers, and prediction templates that need evaluated packs, adapter scripts, and operator-grade run receipts. Primary path for most launchable EarnClaw templates today.

Upstream docs


OpenClaw

What OpenClaw is

OpenClaw is an agent gateway runtime for soul-driven agents: persisted workspace artifacts, skills, and scheduled finance jobs. EarnClaw hosts OpenClaw with managed secrets, volume state, and dashboard access to the gateway when enabled for your org.

Best for

Teams that want OpenClaw-native skills, SOUL-style strategy files, and gateway tooling alongside EarnClaw guardrails. Spot trading, perps, and prediction templates ship on OpenClaw where the catalog marks them launchable.

Upstream docs


Claude managed agents

What Claude managed is

Claude managed agents are Anthropic's hosted agent runtime: model, system prompt, tools, and MCP servers defined in Claude Console, with sessions run in managed sandboxes. EarnClaw wires finance templates to this path when you want Claude-native orchestration without hosting Hermes or OpenClaw yourself.

Best for

Claude-first teams, MCP-heavy workflows, and template lanes validated for managed sessions. Useful when your operators already live in Claude Console and want EarnClaw policy plus execution behind the same agent.

Upstream docs


Quick comparison

FrameworkHosted byTypical lanesOperator surface
HermesEarnClawTrading, yield, prediction packsAgent page, cron diagnostics, run history
OpenClawEarnClawSoul-driven finance agents, skillsGateway artifacts, dashboard access, run history
Claude managedAnthropic + EarnClaw wiringMCP-orchestrated finance templatesClaude Console + EarnClaw agent page

Runtime lifecycle (all frameworks)

Template → Runtime provision → Policy / delegation → Schedule → Preflight → Execute → Observe
  1. Provision — org, agent wallet, runtime id, pack version
  2. Preflight — delegation scopes, template rules, plan allowance, venue readiness
  3. Run — cron, signal, or manual trigger
  4. Receipt — execution status and skip reasons in run history

Health and upgrades

  • Runtime diagnostics surface adapter errors, funding gaps (for example perps margin), and pack version drift
  • You can pin pack versions or roll out upgrades on a schedule that fits your org