Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/archestra-ai/archestra/llms.txt

Use this file to discover all available pages before exploring further.

Agents are the core building blocks of Archestra — reusable AI workers that combine a system prompt, tool access, and optional knowledge retrieval into a single, invokable unit. Once configured, the same agent can be called from the chat UI, triggered by an incoming email, fired on a schedule, reached over HTTP, or embedded inside Slack and Microsoft Teams, without rebuilding the workflow each time.

​
What an Agent Contains

An agent is composed of several optional capabilities that work together at runtime:

System Prompt

Defines the agent’s behavior, persona, and constraints. Supports Handlebars templating so instructions adapt per-user at runtime.

Suggested Prompts

Pre-configured example messages surfaced in the chat UI to help users get started quickly with common tasks.

Tools

One or more MCP tools the agent can call during a run. The full list is exposed via tools/list, or selectively with Load Tools When Needed mode.

Knowledge Sources

One or more Knowledge Bases or connectors that give the agent retrieval access to your internal data without hardcoding sources into the prompt.

Delegation Targets

Other agents this agent can delegate subtasks to, with full call-chain observability and inherited tool guardrail trust state.

​
Invocation Paths

The same agent is reachable through multiple entry points, all managed from Agent Triggers in the sidebar:

Chat UI

Interactive, turn-by-turn conversations directly in the Archestra web interface.

Webhook / A2A

HTTP endpoint following the A2A 1.0 protocol for programmatic or agent-to-agent invocation.

Scheduled Tasks

Cron-based runs that fire automatically and record the full conversation for review.

Incoming Email

Outlook-based email aliases that route incoming messages to the agent and optionally reply.

Slack

Bot integration that routes channel mentions to the agent and replies in-thread.

Microsoft Teams

Azure Bot integration that handles channel mentions and DMs with agent selection per channel.

​
Load Tools When Needed

By default, an agent exposes every assigned tool through MCP tools/list. For large toolsets this can bloat the context sent to the model on every turn. Enable Load tools when needed to keep the initial tool list small. When active, MCP clients see only two built-in tools at first:
  • search_tools — discovers assigned tools by name or description
  • run_tool — executes any assigned tool by name
The full assigned toolset remains accessible; the model just fetches what it needs rather than receiving everything upfront. Tool call policies still apply: if the model uses run_tool to call send_email, Archestra evaluates the send_email policy with the same arguments and context as a direct call.
search_tools and run_tool are enabled implicitly when this mode is on — you do not need to assign them manually.

​
Knowledge Sources

When at least one Knowledge Base or connector is assigned, Archestra automatically adds the built-in query_knowledge_sources tool to the agent. The model calls it during a run to search across all assigned sources and pull relevant context into its answer. Knowledge sources are configured separately and reused across multiple agents. See Knowledge Bases for the full setup guide.

​
Delegation

Agents can delegate subtasks to other agents. Archestra tracks the full call chain for observability across the entire run. Delegated agents inherit the current tool guardrails trust state, so downstream policy enforcement does not reset mid-run.

​
Convert to Skill

Any agent can be converted into an Agent Skill — a reusable SKILL.md instruction set that any other agent can activate via a /slash-command. Use this when the agent’s value is primarily in its instructions and you want them available without requiring users to switch agents. The Convert to skill action opens a confirmation dialog where you:
  • set the skill’s description (required — used by activating agents to decide when to invoke the skill)
  • choose whether to remove the source agent after conversion
Conversion is intentionally lossy — a skill carries instructions only, with no tools, model, or knowledge of its own:
Agent fieldWhat happens
System promptBecomes the skill body
ScopeCarried over directly
NameNormalized to a slug (e.g. Support Helper → support-helper)
DescriptionRequired; prefilled from agent, or generated via LLM
Assigned toolsCarried into the skill’s allowed-tools frontmatter
Default modelNot carried — no skill equivalent
Knowledge sourcesNot carried — no skill equivalent
Suggested promptsFolded into body or metadata
Use Generate in the description field to draft a description from the agent’s prompt, tools, and example prompts via a single LLM call.
If the system prompt uses Handlebars templating, the skill is automatically flagged templated: true so its body is re-rendered with the activating user’s context at runtime rather than baking one author’s values into shared instructions.

​
System Prompt Templating

Agent system prompts support Handlebars templating. Templates are rendered at runtime before the prompt is sent to the LLM, with the current user’s context injected automatically. Agent Skills can opt into the same rendering with a templated: true frontmatter field.

​
Variables

VariableTypeDescription
{{user.name}}stringName of the user invoking the agent
{{user.email}}stringEmail of the user invoking the agent
{{user.teams}}string[]Team names the user belongs to

​
Helpers

HelperOutputDescription
{{currentDate}}2026-03-12Current date in UTC (YYYY-MM-DD)
{{currentTime}}14:30:00 UTCCurrent time in UTC (HH:MM:SS UTC)
All built-in Handlebars helpers (#each, #if, #with, #unless) are available alongside Archestra-specific helpers: includes, equals, contains, and json.

​
Example

You are a helpful assistant for {{user.name}}.
Today's date is {{currentDate}}.

{{#includes user.teams "Engineering"}}
  You have access to engineering-specific tools and documentation.
{{/includes}}

{{#if user.teams}}
  The user belongs to:
  {{#each user.teams}}{{this}}{{#unless @last}}, {{/unless}}{{/each}}.
{{/if}}

Build docs developers (and LLMs) love

Get started for freeTalk to us