HybridClaw

What HybridClaw does

HybridClaw is a TypeScript runtime for running AI assistants inside a company. Its main promise, per the README, is practical business work: bundled skills for CRM, marketing, analytics, finance, cloud, office and operations are written as tested helpers with eval scenarios, credential boundaries and approval tiers, and are validated against Qwen3.6-27B-FP8 as a small-model baseline. Named targets include Salesforce, HubSpot, GA4, Google Ads, Airtable and Microsoft 365.

Credentials stay outside the model context: secrets live in an encrypted store and are resolved at execution time, so the model sees the action and approval context rather than raw keys. Agents can be local, hosted proxy agents or trusted peer instances that exchange end-to-end encrypted A2A envelopes. Other features include a guided first-run onboarding, local MLX model setup on Apple silicon, and an Apps gallery for generated HTML apps and dashboards. The repository carries hermes and hermes-agent topics, but the opening part of the README does not explain a Hermes integration.

Key features

• Business skills for CRM, marketing, analytics, finance, cloud and office tasks
• Encrypted secret store with credentials resolved at execution time
• Approval tiers and sandboxed execution
• Local, hosted and peer agents with encrypted A2A messaging
• Guided first-run onboarding with audit events
• Local MLX model setup on supported Apple silicon Macs

When to use it

• Running CRM and reporting tasks on a small local model with approval gates
• Operating a fleet of role-specific assistants
• Generating internal dashboards and documents as shareable apps

Who it is for: Teams that want a self-hosted, approval-based AI assistant runtime for business operations.

How it fits with Hermes Agent

Tagged hermes and hermes-agent alongside openclaw, but the opening part of the README does not describe how Hermes Agent connects.

FAQ