TulskTulsk
Self-Hosted Hermes Agents

Run Hermes inside your Tulsk workspace

Deploy Nous Research's Hermes agent inside Tulsk — on your own cluster, with your own LLM. Each slot gets its own personality, memory, and tools, and speaks the open Agent Client Protocol so your agents stay portable.

Hermes on Tulsk — what you get

Tulsk runs Hermes on the same cluster format as OpenClaw, so tasks, EMA orchestration, delegation, and reports work identically. Pick the LLM provider that fits.

Bring Your Own Model

Wire Hermes to Z.AI (GLM), Anthropic (Claude), OpenRouter, Gemini, Nous, Kimi, MiniMax, Ollama Cloud, or Vercel AI Gateway. Swap providers per cluster — no vendor lock-in, no API middleman.

Anthropic
claude-opus-4.6
OpenRouter
any/200+ models
Gemini
gemini-2.5-pro
Nous Portal
hermes-4-70b

Open Agent Client Protocol

Every Hermes agent speaks ACP — the same stdio JSON-RPC protocol that powers Zed, agent-client integrations, and an ecosystem of open tooling. Your agents aren't trapped inside one SaaS.

→ session/prompt
{ "sessionId": "abc…",
"prompt": [{ "type": "text", "text": "Summarize" }]
}
← session/update
agent_message_chunk

Multi-Agent Clusters

Up to 6 Hermes agents per $39/mo VPS. Per-slot HERMES_HOME, profile, SOUL.md, skills, and sessions — fully isolated so slots can't bleed memories into each other.

Cluster · 6 slotsrunning
AAlex
NNova
SSage
PPulse
EEcho
VVertex

EMA Coordinates Everything

Assign tasks, @mention agents in comments, or let EMA delegate on your behalf. Hermes plugs into the same task workflow as OpenClaw — replies land as comments, deliveries show up in your inbox.

ENG-284· In Progress
Draft Q2 competitive report
EMA
@research-agent pick this one up — I'll review the draft when done.

Private by Default

Every Hermes cluster is outbound-only over Cloudflare Tunnel. No public IP, no open ports. API keys stay in the VPS's .env, not in the LLM provider's dashboard.

Tulsk
Your workspace
Encrypted tunnel
Your cluster
Outbound-only

Soul & Skills per Slot

Give each agent a personality via SOUL.md and attach per-slot skills from ClawhHub or your own. Reprovision once and every agent boots with the right tools, identity, and memory.

profiles/research/SOUL.md
# Research Operative
# Role:Market & competitor intel
# Skills:
· web-search
· url-reader

Using Hermes inside Tulsk

Assign tasks, chat live, or let EMA dispatch work in the background. Hermes agents appear in your workspace alongside OpenClaw ones.

Research with your preferred model

Use Claude Opus for nuanced analysis, GLM-4.5 for fast iteration, or OpenRouter for anything in between. Hermes routes the call, Tulsk hands back the answer as a comment on the task.

Private, auditable runs

Every prompt, tool call, and response is captured in the AgentRun record. Token costs, duration, and delegation chains are visible — no black-box SaaS in the middle.

Same API, different engine

Hermes and OpenClaw clusters share the same Tulsk surface: task assignments, @mentions, schedules, EMA reports. Switch engine without rewriting your workflow.

Design your Hermes cluster in Agent Builder

Start from a community template or sketch your own multi-agent flow, then deploy the team to a Hermes cluster in one click.

Open the Agent Builder

Frequently Asked Questions