What is HybridClaw
HybridClaw is an enterprise-ready AI coworker that unifies Discord, web, and terminal workflows. It provides a shared assistant brain across these platforms, offering consistent memory, guardrails, and tool execution.
How to use HybridClaw
- Install CLI: Run
npm install -g @hybridaione/hybridclaw. - Initialize workspace: Execute
hybridclaw onboardingto accept the trust model and generateconfig.json. - Start runtime: Launch the core service with
hybridclaw gateway. Usehybridclaw gateway start --sandbox=hostif HybridClaw is already running within a container. - Connect interface: Use Discord or run
hybridclaw tuifor a terminal-native chat experience. - Restart safely: Use
hybridclaw gateway restart [--sandbox=container|host]when configuration or runtime placement changes.
Features of HybridClaw
- Unified Workflows: Seamlessly integrates Discord, web, and terminal environments.
- Shared Assistant Brain: Consistent memory, guardrails, and execution flow across all channels.
- Enterprise-Ready: Built with security, compliance, and operational reliability in mind.
- Local Model Support: Integrates with Ollama, LM Studio, and vLLM providers.
- Activity-Based Timeouts: Extends agent IPC timeouts for ongoing model progress.
- Session Reset Workflow: Clears history and restores defaults for a clean state.
- Runtime Stability: Fixes for worker pools, host-mode prompts, and compaction budgets.
- Trust-First Onboarding: Explicit trust model acceptance required.
- Security & Control: Hash-chained audit trails, runtime guardrails, and sandbox execution.
- Memory Continuity: Persistent workspace memory, semantic recall, and cross-channel session persistence.
- Proactive Scheduling: Schedule recurring or one-shot tasks with cron/interval/at semantics.
- Tool Execution: Built-in capabilities for file operations, shell commands, web fetching, and browser automation.
- Configurable Personality: Switchable agent personas via
SOUL.md. - Observability Export: Structured audit events pushed to HybridAI ingest.
Use Cases of HybridClaw
- Team Productivity: A single AI assistant for research, support, and daily tasks across different communication channels.
- Automated Workflows: Executing tasks and following through on scheduled jobs across the team's digital environment.
- Secure Automation: Leveraging tools and shell commands with built-in safety guardrails.
- Knowledge Management: Grounding AI responses in organizational documents and persistent memory.
FAQ
- Onboarding Trust Model: HybridClaw requires explicit trust-model acceptance before runtime starts, recorded in
config.json. - Configuration: Runtime behavior is managed via typed
config.jsonand secrets in~/.hybridclaw/credentials.json. - Shell Command Safety: Tools execute in ephemeral Docker containers with read-only filesystems and other security measures. Host sandbox mode is available for specific environments.
- Browser Tools: Can test login flows when explicitly requested, with sensitive credential redaction from logs.
- Tool Selection: Defaults to
web_fetchfor static content, escalates to browser automation for dynamic or interactive sites. - Audit Trail: Logs are append-only and hash-chained for tamper-evidence; integrity can be verified with
hybridclaw audit verify. - Instruction File Changes: TUI checks critical instruction files (
SECURITY.md,TRUST_MODEL.md) before startup; drift can be reviewed and synced. - Discord Independence: Can be used without Discord via the
hybridclaw tuicommand. - Heartbeat System: A configurable periodic poll to check for tasks and proactively engage if needed.
- Multi-Channel Bots: Different Discord channels can use different bots, models, and settings.
- Scheduling: Supports chat-created tasks and config-backed jobs with various scheduling options and delivery methods.
- HybridAI Integration: Works with any model available through the HybridAI API, allowing for custom bot creation and configuration.
- Memory Persistence: Remembers things between conversations through daily memory files and a long-term
MEMORY.md. - Personality Switching: Assistant personality can be changed at runtime using the
/personalitycommand. - Skill Compatibility: Supports OpenClaw/CLAUDE-compatible
SKILL.mdfrontmatter.




