Salesforce Agentforce
Type: low-code · Vendor: Salesforce · Language: Apex/Metadata · License: proprietary · Status: active · Status in practice: mature · First released: 2024-09-12
Agentforce is Salesforce's platform for building and deploying autonomous agents that reason over CRM data using topics, instructions, and actions, driven by the Atlas reasoning engine.
Description. Agentforce is Salesforce's enterprise platform for building autonomous agents grounded in CRM data and business processes. Agents are assembled from topics, instructions, and actions, and are reusable through agent templates that produce multiple agents with the same topics and actions. The Atlas reasoning engine drives each agent through a cycle of reasoning, acting, and observing until the user's goal is met. Topics scope which actions are valid and which policies apply at any moment.
Agent loop shape. Each request is classified into a topic that defines which actions are valid and which policies apply. The Atlas reasoning engine then runs a cycle of reasoning, acting, and observing: it thinks through a step, invokes an action such as a Salesforce flow or API, checks the outcome, and loops until the user's goal is complete. Agent definitions are packaged as reusable templates so one template produces multiple consistent agents.
Primary use cases
- autonomous CRM and service agents
- reusing agent definitions across many agents via templates
- topic-scoped action selection over Salesforce flows and APIs
- grounding agents in business data and processes
Key concepts
- Atlas Reasoning Engine → react (docs) — Agentforce's reasoning system that classifies each request into a topic and then loops through reasoning, acting, and observing until the user's goal is met.
- Topics (Subagents) (docs) — Named scopes that categorize the job to be done and define which actions are valid and which instructions and policies apply while that scope is active; renamed subagents in April 2026.
- Actions → tool-use (docs) — The tools an agent invokes to perform tasks, built from Apex, flows, named SOQL queries, or external APIs, with typed inputs and outputs.
- Agentforce Data Library → naive-rag (docs) — A managed knowledge base that ingests, chunks, and indexes Salesforce knowledge articles, uploaded files, and web data so agents can retrieve and ground answers on trusted company data.
- Agent Templates → agent-factory (docs) — Reusable agent definitions that package topics, instructions, and actions so an organization can stamp out multiple consistent agents from one definition.
Patterns this low-code implements —
- ★★ReAct
The Atlas reasoning engine drives each agent through an explicit reason-act-observe cycle: it thinks through a step, takes an action, checks the outcome, and loops until the goal is met, with topics…
- ★★Tool Use
Agentforce actions are the typed tools an agent invokes to perform tasks — built from Apex invocable methods, flows, named SOQL queries, or external APIs exposed through OpenAPI — rather than the mod…
- ★★Naive RAG
The Agentforce Data Library grounds an agent in company data through retrieval-augmented generation: documents are chunked and indexed, and retrieved chunks are added to the prompt so the model answe…
- ★Agent Factory
Agentforce manufactures agent instances from reusable agent templates that package topics, instructions, and actions, deployed via metadata so one template definition produces multiple consistent age…
Neighbourhood
Click any neighbour to follow the lineage. Scroll to zoom, drag to pan.