Why an AI agent glossary matters
AI automation language can be confusing because vendors often use the same words in different ways. A "local AI agent," "workflow automation," "tool call," or "model provider" can mean different things depending on the product, deployment, and permissions involved.
For Hermes Agent, clear definitions help users answer practical questions:
- What can the agent actually do?
- Which tools, files, apps, and APIs can it access?
- Where might prompts, outputs, logs, and data be processed?
- Which workflows should stay human-reviewed?
- Which setup choices matter for Canadian organizations thinking about privacy, operations, and vendor risk?
Starter glossary entries
AI agent
Software that uses an AI model to interpret a goal, reason through steps, and take actions through configured tools. Unlike a basic chatbot, an agent may search files, call APIs, create drafts, schedule tasks, or operate connected systems when permissions allow it.
Local AI agent
An agent that runs on a user-controlled machine or infrastructure rather than only inside a third-party SaaS interface. "Local" can describe where the agent process runs, but it does not automatically mean every model request, API call, log, or file remains local.
Hermes Agent
A local-first AI agent system for automating tasks across configured tools, files, messages, schedules, and workflows. Its capabilities depend on the user's setup, enabled integrations, model provider, permissions, and deployment choices.
Tool calling
The process where an AI agent uses a connected tool to do something beyond generating text. A tool call might search files, send a message, read a calendar, create a document, run a command, or query an API.
Model provider
Supplies the AI model an agent uses to interpret requests and generate outputs. Providers may be cloud APIs, local model runtimes, hosted inference services, or private deployments. Provider choice can affect data location, logging, latency, cost, and vendor review requirements.
Local-first AI
An approach where workflows are designed to run close to the user's own machine, files, and infrastructure whenever practical. Emphasizes user control, inspectability, and direct integration with local tools. Local-first is a design approach, not a legal compliance guarantee.
Data residency
Refers to where data is stored, processed, transmitted, logged, backed up, and accessed. For AI agents, data residency questions can involve prompts, uploaded files, tool outputs, model-provider logs, cloud backups, and message history.
Workflow automation
Uses software to complete repeatable steps with less manual effort. In an AI-agent context, workflow automation may include research summaries, meeting prep, content briefs, invoice follow-ups, document review, or recurring reports.
Scheduled agent
Runs a defined task at a planned time or interval. Examples include a daily inbox summary, weekly analytics brief, monthly client-report draft, or recurring market-monitoring workflow. Scheduled tasks should have logging, error handling, and clear rules for when human review is required.
Autonomous agent
Can pursue a goal with limited step-by-step human instruction. In practice, autonomy should be bounded by permissions, tool access, budget limits, approval gates, and monitoring. Useful agent autonomy depends on careful configuration, not blind trust.
Human-in-the-loop
A person reviews, approves, or guides an agent at important points. This is especially important before publishing content, sending client messages, modifying records, deleting data, making purchases, or acting on sensitive information.
Prompt
The instruction or context given to an AI model. Prompts may include the user's request, system instructions, examples, retrieved documents, tool results, and formatting requirements. Strong prompts define the goal, constraints, desired output, tools to use, and verification expectations.
Context window
The amount of text, tool output, or conversation history an AI model can consider at one time. Larger context windows can help with long documents, but they do not replace good retrieval, summarization, or workflow design.
Retrieval
The process of finding relevant information from files, notes, databases, search indexes, or other sources before an agent answers or acts. Retrieval can help an agent ground its response in the user's actual data. Retrieval workflows should respect file permissions.
MCP
Stands for Model Context Protocol. An integration approach that can let AI systems connect to external tools, data sources, or services through a standardized interface. Protocols like MCP can make tool ecosystems easier to extend, but each connected tool still needs security and permission review.
Webhook
A way for one system to notify another system when an event happens. For example, a form submission, payment event, new support ticket, or repository update could trigger an automation. Webhook-driven workflows need authentication, validation, and safe handling of incoming data.
CLI
Stands for command-line interface. It is a text-based way to run commands, configure tools, and automate workflows from a terminal. CLI workflows are useful for local setup, debugging, scripting, and developer-oriented automation.
API key
A credential that lets software access a service. API keys may grant access to AI model providers, messaging platforms, productivity tools, databases, or other integrations. API keys should be stored securely, scoped narrowly where possible, and rotated when needed.
Secrets management
The practice of storing, accessing, rotating, and revoking sensitive credentials such as API keys, tokens, passwords, and certificates. Agents that can use tools often need credentials. Those credentials should not be hard-coded into public files or casually pasted into prompts.
Vector database
Stores numerical representations of text, images, or other data so related information can be searched by meaning rather than exact keywords. Vector databases are often used for retrieval workflows. Teams should ask where embeddings are generated and stored.
Self-hosting
Means running software on infrastructure you control instead of relying entirely on a vendor-hosted service. Self-hosting can increase control and customization, but it also creates maintenance, security, backup, and monitoring responsibilities.
Data flow
Describes how information moves through a workflow: from user input to agent prompt, tool call, model provider, storage, logs, output, and human review. Mapping data flows helps teams understand exposure and responsibility.
Agent gateway
An access layer that lets users interact with an agent through a channel such as a messaging app, web interface, or internal tool. It may handle routing, authentication, delivery, and notifications. Gateways should be configured with access control and operational logging in mind.
Kanban agent workflow
Uses cards, states, and handoffs to coordinate agent work. Tasks may move from ready to running to blocked or complete, with comments and metadata helping future workers understand what happened. Kanban-style coordination can make longer agent projects more auditable.
Local app automation
Lets an agent operate or assist with applications on the user's machine, such as notes, browsers, files, calendars, or productivity apps. The exact capabilities depend on the operating system, permissions, and available tools.
Bilingual workflow automation
Supports work in more than one language, such as drafting English and French content, preparing translation checklists, or organizing bilingual campaign materials. AI-generated translations and localized content should be reviewed by qualified humans when accuracy matters.
Suggested glossary page template
When each term becomes its own page, use a consistent structure:
- Definition in one or two plain-language sentences.
- Example of the term in a real workflow.
- Why it matters for Hermes Agent.
- Canada-specific note when useful and accurate.
- Related terms and internal links.
- Practical next step, such as reading a setup guide or mapping a data flow.