Documentation

Feature Comparison: Personal Agent vs. OpenClaw, Hermes Agent, Home Assistant

A side-by-side of what each system has. This compares Personal Agent (this repo, a multi-tenant LLM chat + agent SaaS) against three reference systems we researched for the roadmap:

  • OpenClaw: single-operator personal-agent framework (built on Pi/earendil).
  • Hermes Agent: NousResearch’s autonomous agent runtime.
  • Home Assistant: the home-automation platform whose integration/entity/automation architecture we model our integration + entity + automation subsystems on.

Legend: ✓ = has it · ◑ = partial / via add-on / not first-class · ✗ = no · ? = unknown. The comparison is necessarily approximate: the three references move fast and some capabilities live in add-ons or external companions rather than the core.

Conversation & Agent Core

| Feature | Personal Agent | OpenClaw | Hermes Agent | Home Assistant | |—||||| | LLM chat agent | ✓ | ✓ | ✓ | ◑ (Conversation agent) | | Streaming responses | ✓ | ✓ | ✓ | ◑ | | Multi-provider models (OpenAI-compat) | ✓ | ✓ | ✓ | ✓ | | Per-model pricing / cost attribution | ✓ | ◑ | ◑ | ✗ | | Budget caps (user/org/global) | ✓ | ✗ | ◑ | ✗ | | Fallback provider chains / routing | ✓ (provider-diverse) | ◑ | ✓ | ◑ | | In-flight context compression | ✓ | ◑ | ◑ | ✗ | | Regenerate / edit-and-rerun | ✓ | ◑ | ? | ✗ | | Mid-run follow-up (queue while running) | ✓ | ◑ | ? | ✗ | | Rewind / shadow-git undo of agent edits | ✓ | ✗ | ◑ | ✗ | | Conversation rewind (truncate + restore) | ✓ | ? | ? | ✗ | | Fork a chat from any point | ✓ | ? | ? | ✗ |

Tools, Code Execution & Sandboxing

| Feature | Personal Agent | OpenClaw | Hermes Agent | Home Assistant | |—||||| | Tool calling | ✓ | ✓ | ✓ | ✓ (LLM API / intents) | | In-process sandboxed code exec | ✓ (Monty) | ◑ | ✓ (PTC via UDS) | ✗ | | Device/host command execution | ✓ (jailed) | ✓ (SSH/OpenShell) | ✓ (6 backends) | ◑ (shell_command) | | On-demand cloud sandbox containers | ✓ | ◑ (Docker) | ✓ (Modal/Daytona) | ✗ | | Command approval / security modes | ✓ (autonomous/approve/judge) | ◑ (allowlists) | ✓ (approval modes) | ✗ | | LLM “judge” guard over tool calls | ✓ | ✗ | ◑ | ✗ | | MCP client (connect external servers) | ✓ (folder integration) | ◑ | ✓ | ✓ | | MCP server (expose this assistant) | ✓ (/api/mcp) | ? | ✓ | ✓ |

Coding Workspace

| Feature | Personal Agent | OpenClaw | Hermes Agent | Home Assistant | |—||||| | Per-chat coding mode | ✓ | ◑ | ◑ | ✗ | | Editor (Monaco) + PTY terminal | ✓ | ◑ (OpenShell) | ◑ | ✗ | | Fuzzy/patch edits + apply_patch | ✓ | ✓ | ✓ | ✗ | | LSP diagnostics + navigation | ✓ | ✗ | ◑ | ✗ | | Formatters on save | ✓ | ◑ | ◑ | ✗ | | Todo / task tracking in run | ✓ | ✓ | ✓ | ✗ | | AGENTS.md / project instructions inject | ✓ | ✓ (SKILL.md) | ✓ (USER.md) | ✗ | | GitHub integration (PRs/issues/webhooks) | ✓ | ◑ | ◑ | ✗ |

Memory & Knowledge

| Feature | Personal Agent | OpenClaw | Hermes Agent | Home Assistant | |—||||| | Long-term memory (vector recall) | ✓ (pgvector) | ✓ (LanceDB/Honcho) | ✓ (9 providers) | ✗ | | Auto-prefetch relevant memories | ✓ | ◑ | ◑ | ✗ | | Editable memory file (MEMORY.md style) | ◑ | ✓ | ✓ | ✗ | | Document RAG | ✓ | ◑ | ◑ | ◑ (via custom) | | Entity RAG / structured entities | ✓ | ✗ | ✗ | ✓ (registries) | | User modeling (learned profile) | ✓ | ◑ (USER.md) | ◑ (USER.md) | ✗ |

Skills & Extensibility

| Feature | Personal Agent | OpenClaw | Hermes Agent | Home Assistant | |—||||| | Agent skills (capability packages) | ✓ | ✓ (SKILL.md) | ✓ | ✗ | | Skill marketplace / hub | ✓ (curated Git catalogs) | ✓ (ClawHub) | ✓ (agentskills.io) | ◑ (HACS) | | Self-authored / self-improving skills | ✓ | ✓ | ✓ | ✗ | | Integration framework | ✓ (HA-style folder) | ◑ | ◑ | ✓ | | Config-flow setup UI + manifests | ✓ | ✗ | ✗ | ✓ | | Hooks (pre/post events) | ✓ | ✓ | ✓ | ◑ (automations) |

Sub-agents & Orchestration

| Feature | Personal Agent | OpenClaw | Hermes Agent | Home Assistant | |—||||| | Sub-agents / delegation | ✓ (explore/delegate) | ✓ | ✓ (delegate_task) | ✗ | | Parallel fan-out | ✓ | ◑ | ◑ | ✗ | | Best-of-N synthesis | ✓ | ✗ | ? | ✗ | | Background / detached sub-agent runs | ✓ | ◑ | ◑ | ✗ | | Per-sub-agent usage/cost tracking | ✓ | ✗ | ? | ✗ |

Automations & Proactivity

| Feature | Personal Agent | OpenClaw | Hermes Agent | Home Assistant | |—||||| | Scheduled automations (cron/heartbeat) | ✓ (Temporal) | ✓ | ✓ | ✓ | | Event-triggered automations | ✓ (entity events) | ✓ (hooks) | ◑ | ✓ (triggers) | | Static (LLM-free) workflow actions | ✓ | ✗ | ✗ | ✓ (scripts/scenes) | | Blueprints / reusable templates | ✗ | ✗ | ✗ | ✓ | | Commitments / proactive nudges | ✓ | ✓ | ◑ | ◑ (to-do) | | Autonomous goal loop | ✓ (/goal) | ◑ | ✓ | ✗ | | Proactive review / agenda | ✓ | ◑ | ◑ | ✗ |

Communication Channels

| Feature | Personal Agent | OpenClaw | Hermes Agent | Home Assistant | |—||||| | Email triage + HITL reply | ✓ | ◑ | ◑ | ✗ | | Signal | ✓ | ✓ | ✓ | ◑ (add-on) | | WhatsApp | ✓ | ✓ | ✓ | ◑ (add-on) | | Matrix / Zulip | ✓ | ◑ | ◑ | ◑ (add-on) | | Telegram / Slack / Discord | ✗ | ✓ | ✓ | ◑ (add-on) | | Multi-channel unified inbox | ✓ | ✓ | ✓ (gateway) | ✗ | | DM pairing / allowlists | ◑ | ✓ | ✓ | ✗ |

Voice

| Feature | Personal Agent | OpenClaw | Hermes Agent | Home Assistant | |—||||| | STT (speech-to-text) | ✓ | ✓ | ◑ | ✓ (Assist) | | TTS (read aloud, streaming) | ✓ | ✓ (Talk Mode) | ◑ | ✓ (Assist) | | Wake word | ✗ | ◑ | ✗ | ✓ | | Self-hosted voice models | ✓ (speaches) | ◑ (MLX) | ✗ | ✓ (Wyoming) |

Multi-tenancy, Auth & Platform

| Feature | Personal Agent | OpenClaw | Hermes Agent | Home Assistant | |—||||| | Multi-tenant (orgs) | ✓ | ✗ | ◑ | ✗ | | OIDC / SSO | ✓ (Keycloak) | ✗ | ◑ | ◑ (custom component) | | Row-level tenant isolation (RLS) | ✓ | ✗ | ✗ | ✗ | | Groups / teams + shared resources | ✓ | ✗ | ◑ | ◑ (per-entity perms) | | Per-user admin-curated models | ✓ | ✗ | ◑ | n/a | | Server-push events (all windows) | ✓ | ✗ | ? | ✓ |

Clients & Surfaces

| Feature | Personal Agent | OpenClaw | Hermes Agent | Home Assistant | |—||||| | Web app (SPA) | ✓ (Quasar) | ◑ | ◑ | ✓ (Lovelace) | | Desktop app | ✓ (Tauri) | ✗ | ✗ | ✗ | | Android app | ✓ (WebView shell) | ✗ | ✗ | ✓ (companion) | | Mobile push notifications | ✓ | ✗ | ✗ | ✓ | | Slash-command palette | ✓ | ◑ | ◑ | ✗ | | Agent persona (name + “Soul”) | ✓ (Settings field) | ✓ (SOUL.md) | ✓ (IDENTITY.md) | ✗ |

Reading the table

Where Personal Agent leads: multi-tenant SaaS foundations the personal-agent frameworks don’t target: OIDC/Keycloak, row-level tenant isolation, orgs/groups, per-user curated models, budgets, cost attribution, plus a deep coding workspace (LSP, formatters, shadow-git rewind, todo) and the security-mode/judge guard over every tool call.

Where the references lead:

  • OpenClaw / Hermes: a couple of channels Personal Agent doesn’t bundle yet (Telegram/Slack/Discord out of the box) and Hermes’ heterogeneous execution backends (Modal/Daytona/Singularity).
  • Home Assistant: the mature integration ecosystem: blueprints, the registry model (device/area/floor/label), wake-word voice, and a companion mobile app with native sensors.

Near-term gaps worth closing: broader bundled channels (Telegram/Slack/Discord) and automation blueprints (reusable parameterized templates), both of which fit cleanly into existing subsystems (the integration framework, comms channels, automations) rather than needing new islands. (MCP both ways is shipped: Personal Agent is an MCP client via the mcp folder integration and an MCP server at /api/mcp that external clients reach with a Personal Access Token or Keycloak OAuth. A skill marketplace - install from admin-curated GitHub SKILL.md catalogs, agentskills.io-compatible - is built into the Skills view, and an in-instance marketplace lets users publish and adopt agents/skills/workflows.)

Compiled 2026-06 from public docs/source of each reference project. Cells reflect a best-effort reading at that time and may lag upstream changes.