How do I diagnose setup problems quickly?
Run hybridclaw doctor. It checks runtime, gateway, config, credentials, database, providers, Docker, channels, skills, security, and disk state in parallel. Use hybridclaw doctor --fix to apply safe remediations where supported, or --json for CI-friendly output. When the environment checks pass but a specific turn still needs inspection, start or restart the gateway with --log-requests to persist best-effort redacted prompts, responses, and tool payloads in SQLite request_log.
Why does onboarding ask me to accept TRUST_MODEL.md?
HybridClaw enforces explicit trust-model acceptance before runtime starts. Acceptance is recorded in config.json with policy version and timestamp so operators can prove security acknowledgement. For CI or other headless shells, HYBRIDCLAW_ACCEPT_TRUST=true can persist that acceptance automatically before credential checks. Technical runtime controls are documented separately in SECURITY.md.
Do I still configure everything through .env?
No. Runtime behavior lives in typed config.json and runtime secrets live in an encrypted ~/.hybridclaw/credentials.json store. HybridClaw reads the master key from HYBRIDCLAW_MASTER_KEY, /run/secrets/hybridclaw_master_key, or a local owner-only credentials.master.key. If .env exists in the current working directory, supported secrets are imported once for compatibility. If you need the whole runtime state elsewhere, set HYBRIDCLAW_DATA_DIR to an absolute path.
Can HybridClaw call an API without exposing the real key to the model?
Yes. Store the key with /secret set <NAME> <VALUE>, then either reference it in a prompt with <secret:NAME> or bind it to a URL prefix with /secret route add. HybridClaw injects the real header in the gateway-side http_request path and persists redacted tool-call arguments instead of the plaintext secret.
Is it safe to let the agent run shell commands?
By default, tools execute inside ephemeral Docker containers with read-only filesystems, memory caps, dropped capabilities, no-new-privileges, PID limits, and a deny-list of dangerous command patterns. When HybridClaw itself already runs inside a container, operators can switch to host sandbox mode to avoid Docker-in-Docker, trading container isolation for workspace fencing and command guardrails.
Can browser tools test real login flows?
Yes, when explicitly requested by the user for the intended site. Runtime guidance allows authenticated browser-flow testing, while sensitive credential values are redacted from structured audit tool-argument logs.
How do I reuse real browser logins?
Run hybridclaw browser login to open a headed Chromium profile, sign in manually, then close the browser. HybridClaw stores that profile under the runtime data directory and reuses it for later browser automation, including cookie and local-storage based sessions.
How does HybridClaw choose web_fetch vs browser tools?
Default is web_fetch for static/read-only retrieval. HybridClaw escalates to browser automation for JavaScript-heavy apps, auth-gated flows, interaction steps, or when web_fetch reports hints like javascript_required, spa_shell_only, empty_extraction, boilerplate_only, or bot_blocked. Browser snapshots also mark pointer- and click-handler-driven custom controls more reliably, which helps the agent interact with modern app-style UIs.
Can I attach files or point a prompt at repo context?
Yes. The built-in web chat accepts uploads and pasted clipboard files or images before send, and TUI can queue a copied local file or clipboard image with /paste or Ctrl-V. You can also add inline context references such as @file:path[:start-end], @folder:path, @diff, @staged, @git:<count>, and @url:https://example.com/spec so HybridClaw can attach the relevant context directly to the prompt.
Can I extend HybridClaw with plugins?
Yes. HybridClaw supports local plugins with typed manifests, plugin tools, memory layers, prompt hooks, and lifecycle hooks. Valid plugins auto-discover from ~/.hybridclaw/plugins/ and project-local .hybridclaw/plugins/ directories, and you can manage them with hybridclaw plugin list, config, install, reinstall, uninstall, and /plugin reload.
Can I back up or transfer an agent?
Yes. hybridclaw agent export creates a portable .claw archive, inspect validates it without extracting, and install restores the agent on another machine. Archives can also bundle selected workspace skills and home plugins. Legacy pack and unpack aliases still work.
Is the audit trail immutable?
Audit logs are append-only and hash-chained per session, so modifications are tamper-evident. Use hybridclaw audit verify <sessionId> to validate integrity, and hybridclaw audit approvals --denied to review denied actions.
What happens if instruction files are changed?
HybridClaw keeps runtime copies of SECURITY.md and TRUST_MODEL.md under ~/.hybridclaw/instructions/. TUI checks them before startup. Use hybridclaw audit instructions to review drift and hybridclaw audit instructions --sync to restore them from installed sources. Sync actions are audit logged.
What is HybridClaw Cloud?
HybridClaw Cloud lets you run your agent in a fully managed, secured Docker container — no local installation, no infrastructure to maintain. Sign up at
hybridai.one, choose a paid plan or redeem a voucher code, and launch your cloud instance in seconds. It supports the same features, channels, and integrations as the local version.
Can I use it without Discord?
Absolutely. Run hybridclaw tui, use the built-in web chat, or connect Slack, Telegram, Microsoft Teams, iMessage, WhatsApp, or email. The same assistant behavior and context model is shared across supported interfaces, and gateway startup stays available even if one transport is temporarily misconfigured.
How are screenshots returned in Discord?
HybridClaw attaches generated screenshots directly to the reply and keeps the text concise (for example: "Here it is.") instead of exposing internal workspace file paths.
What is the heartbeat system?
A configurable periodic poll (default every 30 minutes) where the agent wakes up, checks its HEARTBEAT.md file for tasks, and proactively reaches out if something needs attention. Silent acknowledgments are discarded to avoid noise.
Can different Discord channels use different bots?
Yes. Each channel is its own session, so it can use different chatbot/model/RAG settings and can also bind to a different agent workspace with /agent switch. You can control reply activation per channel with /channel-mode and guild policy via /channel-policy.
How do I allow free-response in only one Discord channel?
Set guild policy to allowlist using /channel-policy policy:allowlist, then run /channel-mode mode:free in the one channel you want active. Leave other channels unlisted (or set them to off).
How does HybridClaw avoid over-replying in busy channels?
In free-response channels, HybridClaw can selectively stay silent when a peer likely already answered, react instead of replying to short acknowledgments (for example "thanks"), and slow pacing during nights/weekends or long back-and-forth exchanges.
How does scheduling work?
Two layers: chat-created tasks (schedule add) and config-backed scheduler.jobs[]. Both support cron, one-shot at, and interval semantics, and config jobs also support immediate one_shot runs. Config jobs can use agent_turn/system_event actions plus channel/last-channel/webhook delivery, with persisted metadata, review-state tracking for one-shot jobs, and auto-disable after repeated failures.
What is HybridAI?
HybridAI is the bot platform that powers HybridClaw. It provides managed chatbots with document knowledge bases, retrieval-augmented generation (RAG), and multi-model support through an OpenAI-compatible API. You create and configure bots on the HybridAI platform, then connect them to HybridClaw.
What AI models does it support?
HybridClaw supports HybridAI models, OpenAI Codex models (via openai-codex/ prefix), OpenRouter models (via openrouter/ prefix), Hugging Face router models (via huggingface/ prefix), and local backends including Ollama, LM Studio, llama.cpp, and vLLM. Use model list to browse available models and switch per channel or session at any time.
Can I migrate an existing OpenClaw or Hermes Agent home?
Yes. Run hybridclaw migrate openclaw --dry-run or hybridclaw migrate hermes --dry-run to preview the compatible workspace files, config values, model settings, and optional secrets that can be imported into a HybridClaw agent. Add --agent <id> to target a different agent and --migrate-secrets when you want compatible secrets moved into the encrypted runtime store.
Does the agent remember things between conversations?
Yes. The active agent writes daily memory files (memory/YYYY-MM-DD.md) and curates long-term insights into MEMORY.md. By default, direct-message continuity is isolated per channel kind and peer. Operators can opt into linked-identity continuity when verified aliases should share one main session scope. Before session compaction, a dedicated flush turn ensures durable notes are persisted. These files plus SOUL.md are injected into every request.
Can I switch the assistant personality at runtime?
Yes. Use /personality (or /personality list) to view profiles, /personality <name> to activate one, and /personality reset to go back to default. HybridClaw persists the active persona in SOUL.md as a managed block with Name, Definition, and Rules.
Is there a web-based admin interface?
Yes. The gateway serves an embedded admin console at /admin with pages for Dashboard, Terminal, Gateway, Sessions, Jobs, Bindings, Models, Scheduler, MCP, Audit, Skills, Plugins, Tools, and Config. The Skills page can create local skills from a form or ZIP upload and review adaptive-skill amendments. There is also a live browser terminal at /admin/terminal, a web chat at /chat, and an agent/session dashboard at /agents. When WEB_API_TOKEN is unset, localhost access opens without a login prompt. When it is set, the auth callback can store the token in browser storage and redirect to a safe relative path after login.
Can I use skills from CLAUDE.md or OpenClaw?
Yes. HybridClaw supports OpenClaw/CLAUDE-compatible SKILL.md frontmatter, including always, requires, and metadata fields. Skills can be loaded from workspace, project .agents/skills, personal (~/.codex/~/.claude/~/.agents), bundled, or enterprise extra directories; higher-priority tiers override lower ones. Community skills can also be imported from packaged official/<skill> sources, skills-sh, clawhub, lobehub, claude-marketplace, well-known, and explicit GitHub repo/path refs. Trust-aware scanning blocks risky personal/community skills while allowing vetted bundled skills.
Powered by HybridAI