This guide is built for first-time operators. It explains the full control-plane flow in plain English: choose the right installer, create a workspace, pair the local runtime, enter LLM runtime values during installation, generate a workflow-aware project from blueprint, deploy to a node, run real work, review records and outputs, capture learnings, and dispatch approved deliveries.
Hermes ships multiple installer paths so operators can start from the public download catalog or from the paired Local Runtime workflow.
Illustrative product screenshot showing the three operator entry points before installation.
Best when the operator needs the installer before a workspace exists or before sign-in.
Use when the runtime should live on an external disk or removable operational kit.
Illustrative account creation surface showing the exact fields a first-time user expects.
From the login page, use Create account. Enter the workspace name, your operator name, email address, timezone, and password.
Hermes requires email verification before the workspace can operate. This keeps runtime pairing and secret management tied to a verified operator identity.
Illustrative local runtime screenshot showing the pairing code, package selection, and approval model.
In Local Runtime, create a pairing session for the correct operating system, architecture, and install mode.
Start the downloaded bootstrapper, enter the pairing code, and follow the guided setup for Hermes, LocalConnector, runtime directories, and LLM runtime values such as API key, base URL, and primary model.
When the node phones home, approve it in Hermes Control Center so the bootstrap manifest can be released.
After installation, the machine should appear under Nodes as online and available for deployment pickup.
For a first-time team, Project Builder is the primary creation flow. The blueprint description becomes the source that generates workflow type, agent topology, prompts, records, outputs, reviews, and delivery surfaces automatically.
Illustrative Project Builder flow showing how Hermes turns the business description into an operating blueprint.
Open Project Builder and enter the business description, workflow type, target outcomes, review expectations, and delivery intent.
Describe the operating goal, audience, review style, artifact expectations, and delivery pattern rather than implementation details. Hermes uses this context during blueprint generation.
The system uses the blueprint description to generate the first project structure instead of requiring you to hand-build every surface in advance.
Deployment should only happen after the blueprint has successfully generated the project and materialized agents, topology, prompt drafts, and schedule.
When the LocalConnector node is online, use Generate blueprint. The system design task runs through Hermes locally and should create the project and materialize its operating structure automatically.
Project Builder should move from a plain-English operating description to an actionable blueprint.
After a healthy blueprint, the project should have active agents, topology links, prompt drafts, and schedule.
Those business artifacts appear only after deployment and execution.
Illustrative control-plane screenshot showing deployment publishing and runtime execution flow.
Open Deployments, pick the project, choose an online node, then publish the deployment package.
The deployment status should move to Applied. That means the node has materialized the runtime profile locally and is ready to execute work.
Keep the LocalConnector online. The node should heartbeat normally and the deployment queue should no longer show the package as pending pickup.
If the deployment is not applied yet, the run will not have the correct local profiles, prompts, or active agents.
Use Run now from the Runs page to create a manual execution wave. Hermes seeds a first supervisor task, then creates downstream work as the run develops.
A healthy run expands beyond one coordination task and eventually produces workflow records, outputs, review items, deliveries, or learnings across multiple pages.
A healthy task life cycle normally moves from Queued to Claimed to In Progress and finally to Completed or Failed.
The first run often begins with one supervisor coordination task. More tasks appear when the supervisor creates work for workers and reviewers.
A strong run should usually create additional tasks after the supervisor task. If the run completes with only one task and no outcomes, inspect prompts, blueprint quality, or local execution errors.
Runs show orchestration progress. Tasks show the individual unit history, retries, and agent-specific execution path.
After a successful run, move page by page to confirm that Hermes produced usable business outcomes rather than only a technically completed run.
Look for reusable findings, entities, risks, blockers, opportunities, source notes, and workflow-specific records. This confirms the run produced auditable operating memory.
Review generated reports, recommendations, briefs, customer-facing drafts, action plans, or other deliverables for quality and completeness.
Inspect review-required outputs, external-facing copy, and sensitive recommendations before anything moves downstream.
A run marked Completed should usually correspond to non-empty records, outputs, approvals, deliveries, outreach artifacts, or learnings. Empty outcomes usually signal a quality or configuration gap.
Illustrative operator screenshot showing the governance phase after Hermes has produced work.
Review structured records and outputs first. For outreach projects, also review discovered leads and add contacts before dispatching email deliveries.
Approve strong recommendations or external-facing messages, reject weak duplicates, and request regeneration when evidence, personalization, or compliance needs work.
Dispatch approved Email, Telegram, reporting, or notification deliveries. Use the delivery status, error badges, and channel diagnostics to guide action.
Approve only the learnings that should shape the next deployment package. The learnings surface is an operator memory layer, not a raw log dump.
Use the dedicated setup guide for the two pieces new operators most commonly need help with: LLM runtime values during installation and Telegram delivery credentials in Settings.
Learn how to create a bot token with BotFather, resolve the right chat ID or public username, store TELEGRAM_BOT_TOKEN in Settings, and validate the target before dispatching.
Learn where to retrieve the API key, base URL, and exact model name for OpenAI, Azure OpenAI, Claude, Gemini, Groq, Mistral, DeepSeek, xAI, OpenRouter, Ollama, and custom OpenAI-compatible gateways, then enter those values during Local Runtime installation.