# Salesforce Agentforce — Salesforce's platform for building, deploying and monitoring AI agents on top of CRM data. > Source: The Agents Index — https://theagentsindex.com/salesforce-agentforce (structured, researched, re-verified) > Facts last verified: 2026-09-03 Agentforce is Salesforce's platform for building AI agents that answer questions and take actions inside its CRM, using the Atlas Reasoning Engine to plan each step. It spans customer service, sales, field service and employee support, plus a coding agent, Agentforce Vibes. | Fact | Value | | --- | --- | | Website | https://www.salesforce.com/agentforce/ | | Price | from $5/user per month | | Free tier | Yes | Salesforce Foundations is a $0 add-on activated inside an existing Salesforce org (not a time-boxed trial). It includes Agentforce Builder, Agent Script, Prompt Builder, Agentforce Coworker and Service/Sales/Employee Agents. Agentforce usage consumes credits: the FAQ states a 200K Agentforce Flex credit grant while the Agentforce feature card states 445k-450k Flex Credits shared with Data 360 - the page publishes both figures. Overage fees apply and usage is tracked in Digital Wallet. Other caps include 2,000 marketing emails per org per month, and a footnote marks the credit inclusion as U.S. only with restrictions. ## What it does **Writes code** · Yes — The Agentforce Vibes coding agent turns a natural-language description into Salesforce code and metadata. > Transform ideas into contextual and structured code using natural language. > — https://www.salesforce.com/agentforce/developers/vibe-coding/ · Capability section, 'Natural Language to Code' block **Debugs and fixes** · Yes — Vibes runs the test or deployment, reads the failure and iterates toward a fix rather than stopping at the error. > Iterates on failures. When a test fails or a deployment errors, it diagnoses the problem and attempts a fix rather than presenting an error and stopping. > — https://developer.salesforce.com/docs/platform/agentforcevibes/guide/afv-agentic-dev.html · 'What Is Agentic Development', 'Iterates on failures' bullet **Writes tests** · Yes — Apex test classes and LWC Jest tests are generated alongside the code, with a dedicated QA Validator sub-agent for coverage. > Test class generation is time-consuming but well-suited to agent execution. QA Validator in Teams Mode handles assertions, bulk test data, and governor limit coverage. > — https://developer.salesforce.com/docs/platform/agentforcevibes/guide/afv-agentic-dev.html · 'Where Agentic Development Fits', 'Test coverage is mechanical but required' bullet **Reviews code** · Partly — Static analysis with auto-fix runs over code in your own workspace and Git diffs can be attached as context, but no diff or pull-request review feature is documented and the docs state the trust layers do not replace code review. > Scan large amounts of code, catch potential bugs, identify runtime efficiency errors, catch security vulnerabilities early in the development process, and automatically fix them > — https://www.salesforce.com/agentforce/developers/vibe-coding/ · Capability section, 'Code Analyzer Integration' block **Whole-codebase work** · Partly — Parallel sub-agents coordinate multi-file, cross-layer changes in isolated Git worktrees, but the docs warn that context truncates, advise preferring specific files over entire folders, and claim no repository-wide index. > Each sub-agent operates in an isolated context and uses distinct Git worktrees to eliminate workspace conflicts while building out features across the Salesforce stack. > — https://developer.salesforce.com/docs/platform/agentforcevibes/guide/afv-overview.html · Capability bullet, 'Autonomous Sub-Agents & Teams' **Ships it** · Yes — It deploys metadata to the org itself, and through the Salesforce DX MCP Server's DevOps Center tools it commits and opens pull requests; deploy and delete stay approval-gated by default. > It creates files, modifies code, runs commands, and deploys metadata directly in your local workspace. > — https://developer.salesforce.com/docs/platform/agentforcevibes/guide/afv-agentic-dev.html · 'What Is Agentic Development', 'Acts in your environment' bullet **Done-for-you service** · Partly — Agentforce is software the customer configures in a low-code builder; Salesforce separately sells Professional Services that performs the hands-on build, reached through an account executive rather than bundled with the product. > while we handle the hands-on build, integration, and data work > — https://www.salesforce.com/services/professional-services/ · 'Get tailored solutions for every step of your Salesforce journey' accordion, 'Build' panel body ## How much it does unattended **Runs autonomously** · Yes — Autonomy is a per-agent setting rather than a fixed trait; agents escalate to a human when an issue exceeds their scope. > Agents can be autonomous (act without human intervention) or can be guided to involve a human in the loop. > — https://www.salesforce.com/agentforce/guide/ · 'Agentforce Fundamentals' section, 'Key capabilities of AI agents' bullet list **Multi-agent** · Yes — A primary agent delegates to secondary specialist agents routed by the Atlas Reasoning Engine, and can delegate to third-party agents over Agent2Agent. > Multi-Agent Orchestration allows you to build a team of specialized AI agents that work collaboratively to solve complex problems. > — https://www.salesforce.com/agentforce/multi-agent-orchestration/ · Multi-Agent Orchestration FAQ, 'What is Multi-Agent Orchestration?' answer **Agent permissions** · Partly — Guardrails and filters bound which topics and actions an agent may reach and data access honours existing Salesforce permissions, but an approval-before-every-action mode is documented only for the Agentforce Vibes coding agent. > Agentforce Guardrails combine user-defined safeguards and Salesforce managed protections to foster trust and maintain control over agent behavior. > — https://www.salesforce.com/agentforce/how-it-works/ · 'Trusted AI, built-in' section, 'Agent Guardrails' card ## Which models it runs on **Claude** · Yes — Anthropic Claude models are named rows in the supported-model table, and the non-default AWS-Hosted Agentforce model option is a Claude model on Amazon Bedrock. > Anthropic and Amazon models are operated on Amazon Bedrock infrastructure entirely within the Salesforce Trust Boundary. > — https://developer.salesforce.com/docs/ai/agentforce/guide/supported-models.html · Standard configuration models table, 'Salesforce Trust Boundary' footnote **GPT** · Yes — OpenAI and Azure OpenAI GPT models fill most of the supported-model table, and the out-of-the-box Salesforce Default option is a GPT-based managed mix. > The Salesforce Default model option for Agentforce is a managed mix of trusted models (currently including GPT-4o) optimized by Salesforce for accuracy, trust, and performance. > — https://developer.salesforce.com/docs/ai/agentforce/guide/supported-models.html · Prose above the standard configuration models table **Open models** · Partly — The Salesforce-managed model list names only Anthropic, OpenAI/Azure OpenAI, Google Vertex, Amazon and NVIDIA (beta), with no Llama or Mistral, so open-weight models arrive only through the customer's own BYOLLM connection. > you can connect the Einstein AI Platform to any language model, including custom-built models > — https://developer.salesforce.com/docs/ai/agentforce/guide/supported-models.html · 'BYOLLM Open Connector' section, first paragraph **Model choice** · Yes — The customer selects the model option in Setup and can override it per agent or subagent in Agent Script; leaving the default hands the choice to Salesforce's managed mix. > To change the default model option for Agentforce, go to the Agentforce Agents page in Setup. > — https://developer.salesforce.com/docs/ai/agentforce/guide/supported-models.html · Prose above the standard configuration models table, first sentence ## Where you use it **In your editor** · Yes — Agentforce Vibes ships as a VS Code extension and is preinstalled in the Salesforce-hosted Agentforce Vibes IDE. > This secure, AI-powered developer tool is available as an easy-to-install VS Code extension. > — https://developer.salesforce.com/docs/platform/agentforcevibes/guide/afv-overview.html · Overview, second paragraph **On the command line** · Partly — Salesforce CLI drives the whole agent lifecycle from a terminal, but the Vibes coding agent itself is distributed only as a VS Code extension and a hosted IDE. There is no standalone Vibes CLI. > Scaffold, deploy, preview, and test agents from the terminal or a continuous integration (CI) system. > — https://developer.salesforce.com/docs/ai/agentforce/guide/headless-tools.html · 'lower-level tools' bullet list, Salesforce CLI entry **In your pipeline** · Partly — Agent tests run in a CI system via Salesforce CLI and DevOps Center tools handle pull requests and pipeline promotion, but Salesforce documents no agent that runs on a pull request by itself. > Add the tests to a continuous integration (CI) system to make sure that your agent continues to work correctly in the future. > — https://developer.salesforce.com/docs/ai/agentforce/guide/agent-dx-test.html · 'Test an Agent with Agentforce DX', testing-flow bullet list, final bullet **In a browser** · Yes — Agents are built and administered in the browser inside Salesforce, and can be embedded on the customer's own website via the Agent API and messaging channels. > The Agentforce Builder is located inside Agentforce Studio in the App Launcher. Access is granted by admin permissions. > — https://www.salesforce.com/agentforce/agent-builder/ · AI Agent Builder FAQs, 'Who has access to Agentforce Builder?' answer ## Whose machine it runs on **Self-hosted** · No — Agentforce runs on Salesforce's own public-cloud infrastructure, and even default model inference is operated inside the Salesforce Trust Boundary; no on-premises or customer-VPC deployment is offered. > Hyperforce is the next-generation Salesforce infrastructure architecture, built for the public cloud. > — https://www.salesforce.com/platform/public-cloud-infrastructure/ · Data Residency FAQ, 'What is Hyperforce?' answer **Open source** · Partly — The Agentforce platform is a hosted Salesforce service and is not distributed as source; the Salesforce Skills Library that grounds Agentforce coding agents is published Apache-2.0 on Salesforce's own GitHub org. > Apache License Version 2.0 Copyright (c) 2026 Salesforce, Inc. All rights reserved. > — https://github.com/forcedotcom/sf-skills/blob/main/LICENSE.txt · LICENSE.txt, opening lines ## What it costs to run **How it meters** · Yes — Consumption is metered per action against a Flex Credit pool: 20 credits per Agentforce action, 30 per voice action, 100k credits for $500. Per-user licensing and a flat $2 per conversation are the alternatives. > Each action is metered individually, so your investment aligns to the value delivered, not just the number of interactions. > — https://www.salesforce.com/agentforce/pricing/ · 'What are Flex Credits and how are they used?' section, third paragraph **Free tier** · Yes — Salesforce Foundations is a $0 add-on activated inside an existing Salesforce org, seeded with a fixed Flex Credit grant and subject to overage fees; every paid Agentforce tier routes to a sales conversation. > Yes, Foundations is a $0 add-on that gives your org access to more functionality. > — https://www.salesforce.com/crm/foundations/ · Salesforce Foundations FAQ, 'Is Foundations really free?' answer **API access** · Yes — The REST Agent API needs Agentforce enabled with an activated agent, is not supported for the 'Agentforce (Default)' agent type, and its usage draws down credits. > With Agent API, you have access to your goal-oriented, autonomous AI agents from anywhere that can call a REST API. > — https://developer.salesforce.com/docs/ai/agentforce/guide/agent-api.html · Agent API Developer Guide, opening paragraph **MCP server** · Yes — Salesforce hosts MCP servers so external agents can query an org under per-user OAuth, and Agentforce is also an MCP client consuming third-party servers attached in Agent Builder. > any compatible AI client—Claude, ChatGPT, Cursor, or a custom agent—can connect to a Salesforce org and act on behalf of authorized users > — https://developer.salesforce.com/docs/platform/hosted-mcp-servers/guide · Salesforce Hosted MCP Servers, opening paragraph **Bring your own key** · Partly — Own inference via BYOLLM from four providers with your own credentials, but requests still route through Salesforce's Models API; own encryption keys require the Salesforce Shield add-on, and the agent runtime always stays on Salesforce. > With BYOLLM, you can add a foundation model from a supported provider, configure your own instance of the model, and connect to the model using your own credentials. > — https://developer.salesforce.com/docs/ai/agentforce/guide/supported-models.html · 'Bring Your Own LLM' section, first paragraph ## Buying it for a team **A company can buy it** · Yes — Sold to organisations through an account executive: every paid tier's call to action is 'Contact us', and only the $0 Foundations add-on is self-serve inside an existing org. > Contact your Account Executive to learn more about what model might work best for your business and use case. > — https://www.salesforce.com/agentforce/pricing/ · Agentforce FAQ, 'How do I know if Flex Credits or Conversation pricing is the right model for my business?' answer, final sentence **Seat model** · Yes — Per-user options are $5/user/month (metered, requires Flex Credits), a $125 add-on, a $150 Industries add-on and Agentforce 1 Editions from $550; no seat minimum or maximum is published. > Choose from consumption-based pricing, with Flex Credits or Conversations, or per-user licensing — with options designed to fit how and where you deploy AI. > — https://www.salesforce.com/agentforce/pricing/ · Page marquee, paragraph under the H1 'Agentforce Pricing' **Pooled budget** · Yes — One fungible org-level credit balance spans teams and channels, with Agentforce 1 Editions bundling 2.5M credits per org per year, and Digital Wallet tracks it and automates internal chargebacks. > Flex Credits can be applied across teams, channels, and use cases, allowing you to expand your AI strategy without renegotiating pricing or forecasting volume. > — https://www.salesforce.com/agentforce/pricing/ · 'What are Flex Credits and how are they used?' section, third paragraph **Admin controls** · Yes — Admins author the agent's topics, actions and guardrails; the Trust Layer additionally masks PII/PCI and filters toxic output, and agent data access honours existing permissions and sharing rules. > Agent Builder enables Salesforce admins and developers to use natural language to create jobs to be done, instructions, actions, and guardrails for their agents. > — https://www.salesforce.com/agentforce/how-it-works/ · Agentforce FAQ, 'Can I customize Agentforce workflows?' answer, final sentence **Audit log** · Yes — Agent actions and outputs are recorded for governance, and Observability adds per-turn session traces; event-level logging with SIEM streaming and Field Audit Trail are the paid Salesforce Shield add-on. > the Audit Trail feature provides the data you need to track AI agent actions and outputs, ensuring AI usage complies with your organization's security, privacy, regulatory, and AI governance policies. > — https://www.salesforce.com/agentforce/how-it-works/ · Agentforce FAQ, 'How secure is Agentforce?' answer, second paragraph **Single sign-on** · Yes — Both protocols are supported, and the org can act as identity provider, service provider or both; Single Logout is configured separately. > Salesforce supports SSO with SAML and OpenID Connect. > — https://help.salesforce.com/s/articleView?id=xcloud.sso_about.htm&type=5 · Single Sign-On article, body prose on selecting the authentication protocol ## What happens to your code **Opt out of training** · Yes — There is nothing to opt out of: the Einstein Trust Layer applies zero retention by default, so prompts and responses are not stored by or used to train the third-party models. > Zero data retention is a strict policy where the prompts and generated responses are never stored or used to train the underlying third-party large language models > — https://www.salesforce.com/artificial-intelligence/trusted-ai/ · 'Zero Data Retention and Data Masking' section, second sentence **Data residency** · Yes — Region is selectable on Hyperforce, with in-country storage conditional on the service being available in that country; strict EU storage and processing with in-region support is the paid Hyperforce EU OZ. > Select global regions on where your Salesforce applications will run with options for the U.S., U.K., Germany, India, Japan, and more. > — https://www.salesforce.com/platform/public-cloud-infrastructure/ · Page marquee, first paragraph under 'Hyperforce: Public Cloud Infrastructure' **Getting out** · Partly — Data is exportable during the term and returnable for 30 days after termination, though insights, reports and scoring may be manual-export-only; unused Flex Credits do not roll over between subscription terms. > During the contract term, customers may export a copy of any Customer Data made available for export through the Covered Services. > — https://www.salesforce.com/en-us/wp-content/uploads/sites/4/documents/legal/misc/einstein-platform-security-privacy-and-architecture.pdf · 'Return of Customer Data' section, first sentence **Certifications** · Yes — The Agentforce-scoped audits section lists SOC 2, ISO 27001/27017/27018, ISO 42001, EU/UK Binding Corporate Rules and Data Privacy Framework; a footnote excludes Agentforce Vibes and Agentforce for Scale, and FedRAMP sits with Government Cloud Plus. > Salesforce has achieved ISO 27001/27017/27018 certification for its ISMS from an independent third party. > — https://www.salesforce.com/en-us/wp-content/uploads/sites/4/documents/legal/misc/einstein-platform-security-privacy-and-architecture.pdf · 'Audits and Certifications' section, ISO 27001/27017/27018 bullet ## Platform record (as of 2026-08-30) Reported by the code-hosting platform, not by the vendor. | Fact | Value | | --- | --- | | Licence | Apache-2.0 | | Stars | 81 | | Forks | 22 | | Last push | 2026-06-02 | ## Provenance 2 of 29 sources behind the facts above are independent of the vendor. ## Coverage elsewhere Pages about this product that the vendor does not control. The label states how independent each one is. | Source | Kind | Read | | --- | --- | --- | | eweek.com | Independent source | 2026-08-23 | | default.com | Independent source | 2026-08-23 | | cmswire.com | Independent source | 2026-08-23 | ## What changed - 2026-09-03 — Capabilities · Commercial terms · Free tier · Price - 2026-08-31 — Capabilities · Commercial terms · Free tier · Price - 2026-08-23 — Alternatives · Documentation link · One-liner · Repository · Summary - 2026-08-19 — Summary ## Correcting this record Anything marked "Not established" is a question we have not answered — it is not a No. If you represent Salesforce Agentforce, or anything here is wrong or out of date, you can supply it with a source: the operations this site accepts from an agent are listed at https://theagentsindex.com/.well-known/agents.json — that catalogue needs no key to read, and each entry states its own auth. Supplying evidence is free. Nothing that can be bought here changes what a field says, whether a gap is closed, or how this record is scored.