Check your company context
AI agent interaction platform

Turn your website into an interface for AI agents

Your next visitor may be an AI agent researching a product, comparing options or acting for a buyer. Salespeak gives those agents a structured way to discover what your company offers, obtain current answers, and complete the actions you choose to support.

Get a demo → See a live agent endpoint

Shipped Discovery and governed answers, through the agent endpoint and MCP server. Roadmap Supported actions. Not available today.

From human-only website to dual interface

Your website keeps serving people. Agents get a door built for them.

An AI agent reading a website has to crawl pages built for human eyes and infer the answer. An agent-facing layer lets it ask a question and get a structured answer instead. The visual website stays exactly where it is.

Definition

An AI agent interaction platform gives external AI agents a deliberate, machine-readable way to reach a company: discover what it offers, get answers grounded in information the company provides, and, where the company chooses, invoke specific workflows under its authentication and policy rules. It runs alongside the website. It does not replace it.

For people

The visual website

Pages, navigation, design and forms stay as they are. Buyers who visit still read, compare and contact you the way they do today.

For software agents

An agent-facing layer

An endpoint an agent can query in natural language, answered from the knowledge sources your team connects. Agents that do not use it keep reading your pages, which is the job of Agent Optimizer.

Three capabilities

Discovery, answers, actions. Two ship today. One is on the roadmap.

Each capability below says what exists now and where the boundary is. Examples are illustrative.

01Shipped

Structured discovery

Agents can understand available products, resources, and supported operations.

Today

Each workspace gets an NLWeb agent endpoint that answers who queries and supports discover. A public discovery API lets an agent check whether a domain has a Salespeak agent and start a conversation with it. Salespeak's own site publishes an MCP server card at /.well-known/mcp/server-card.json.

Boundary

The operations exposed today are queries. No action operations are exposed.

Example

A procurement agent shortlisting data-security vendors checks which vendor domains have an agent, then asks each one who it represents before reading any pages.

02Shipped

Governed answers

Agents receive current answers grounded in approved company context.

Today

The agent endpoint answers ask queries in three modes (list, summarize, generate) with Schema.org responses, grounded in the knowledge sources your team connects: your website, uploaded files, Q&A added in the app, and a Google Drive folder. Salespeak's own endpoint returns the source URLs each answer drew on.

Boundary

Salespeak governs what the endpoint answers from. It does not control what an external AI model does with the answer after it receives it.

Example

An IT buyer's agent asks whether a vendor integrates with its CRM and gets an answer drawn from the integration documentation the vendor connected, instead of piecing one together from third-party pages.

03Roadmap

Supported actions

Agents can invoke only explicitly exposed workflows, subject to authentication, authorization, and policy.

Today

Not available. Salespeak does not offer production actions for external agents. The definition above is the design principle actions will follow when they ship.

Boundary

Action confirmation, authorization rules, audit trail and human review will be defined before any action ships.

Example (vision)

With the buyer's authorization, an agent requests a demo with a vendor's sales team, through a workflow the vendor chose to expose.

Example buyer-agent journey

One evaluation, five steps. Each labeled with what ships.

An illustrative scenario. Steps 2 to 4 describe what the agent endpoint supports today. Step 5 is vision, not product behavior.

  1. 1

    A buyer asks their agent to evaluate vendors

    Buyer side

    A security lead asks an AI assistant to shortlist vendors for a compliance tool. This step happens outside Salespeak.

  2. 2

    The agent discovers the vendor's agent

    Shipped

    The agent uses the public discovery API to learn that a vendor's domain has a Salespeak agent, then sends a who query. This happens only if the buyer's agent is built to look for an endpoint.

  3. 3

    It retrieves product, integration, security and pricing facts

    Shipped

    It sends ask queries and receives structured answers from the vendor's knowledge sources. The answers are only as complete and current as those sources.

  4. 4

    It asks a governed follow-up question

    Shipped

    It asks a narrower question in the same conversation, such as which plan includes single sign-on, and gets an answer grounded in the same sources.

  5. 5

    With buyer authorization, it requests a demo

    Vision

    The agent invokes a demo-request workflow the vendor exposed, after the buyer approves. Not available today. See architecture and trust for what must exist first.

Architecture and trust

Where answers come from, and who can reach them.

Company knowledge sources feed a governed context layer. External agents reach it through the agent endpoint; your team's own AI tools reach it through an authenticated MCP server. Actions are drawn dashed because they do not exist yet.

Agent Interaction Platform architecture Company knowledge sources (website crawled daily, uploaded files, Q&A added in the app, Google Drive folder) feed the Salespeak context layer. Governed answers are shipped; supported actions are roadmap and drawn dashed. Answers reach external AI agents through the agent endpoint (NLWeb who, ask and discover, plus a public discovery API) and reach your team's AI tools through a per-workspace MCP server with OAuth 2.1. A dashed roadmap line runs from supported actions to external AI agents. Company knowledge sources SHIPPED Website, crawled daily Uploaded files Q&A added in the app Google Drive folder Salespeak context layer Governed answers SHIPPED Grounded in your knowledge sources Supported actions ROADMAP Exposed workflows only. Not available today. Agent endpoint NLWeb: who, ask, discover Public discovery API SHIPPED MCP server OAuth 2.1 One per workspace SHIPPED External AI agents Research and buyer agents Your team's AI tools Any MCP client Shipped Roadmap
Actions are not part of the shipped architecture. The dashed path shows where they would sit.
Source of truth Shipped
The knowledge sources your organization connects: your website, uploaded files (PDF, Word, PowerPoint, Excel, CSV, text, Markdown), free-text Q&A added in the app, and a shared Google Drive folder. Answers are grounded in those sources.
Freshness Shipped
The website is crawled daily, respecting your sitemap and noindex. Google Drive changes sync on a schedule. Files and Q&A change when your team edits them in the app.
Permissions and identity Shipped
The per-workspace MCP server requires OAuth 2.1. The agent endpoint is built for external agents, and Salespeak's own public endpoint requires no authentication, so treat anything it can answer from as public information.
Privacy and security
Salespeak's security program and trust center are documented on the security page.
Action confirmation Roadmap
No actions are exposed today, so there is nothing to confirm. Buyer authorization and confirmation rules will be defined before actions ship.
Audit trail Roadmap
Applies to actions. The audit record for agent-invoked workflows will be defined before actions ship.
Failure behavior Roadmap
Applies to actions. What happens when an action fails or is refused will be defined before actions ship.
Human review Roadmap
Applies to actions. Which workflows require a person to approve them will be defined before actions ship.
FAQ

Questions technical buyers ask

What is an AI agent interaction platform?

Software that gives external AI agents a deliberate, machine-readable way to reach a company: discover what it offers, get answers grounded in information the company provides, and, where the company chooses, invoke specific workflows under its authentication and policy rules. Salespeak ships discovery and answers today. Actions are on the roadmap.

How is this different from a chatbot?

A chatbot talks to a person in a window on your website. An agent interaction platform serves software agents through endpoints that return structured responses, with no widget involved. Salespeak's product for human visitors is the Website Inbound Agent, and it is not agent-to-agent interaction.

How is it different from an API or MCP server?

MCP and NLWeb are protocols, and an MCP server or API is one interface. The platform is the governed layer behind the interfaces: the knowledge sources, how answers are grounded in them, who can reach which interface, and, once actions ship, the policy for them. Salespeak exposes that layer through a per-workspace NLWeb agent endpoint and a per-workspace MCP server with OAuth 2.1.

Can an external agent take actions automatically?

Not today. Supported actions, such as an agent requesting a demo for a buyer, are on the roadmap. When they ship, agents will be able to invoke only workflows you explicitly expose, subject to authentication, authorization and policy.

How are answers governed and kept current?

Answers are grounded in the knowledge sources your team connects: your website, uploaded files, Q&A added in the app and a Google Drive folder. The website is crawled daily, Drive changes sync on a schedule, and files and Q&A change when your team edits them. Salespeak does not control what an external AI model does with an answer after it receives it.

What prevents unauthorized actions?

Today, no actions are exposed to external agents, so none can be invoked through Salespeak. The per-workspace MCP server requires OAuth 2.1. Authorization, confirmation and audit rules for actions will be defined before actions ship.

Does this replace our website?

No. People keep using the visual website. The agent-facing layer is an addition for software agents, and agents that do not use it keep reading your pages.

Which protocols and agents are supported?

Each workspace gets an NLWeb agent endpoint answering who and ask queries with Schema.org responses, and a public discovery API lets an agent check whether a domain has a Salespeak agent. Each workspace also gets a Model Context Protocol server with OAuth 2.1, usable from Claude, ChatGPT, Cursor or any MCP client. Salespeak's own endpoint publishes an MCP server card following the SEP-1649 draft; see how it works. Whether a given AI agent uses these interfaces depends on that agent. Neither protocol is universally adopted.

Talk it through

Decide how agents should reach your company.

We will walk through what the agent endpoint can answer from your knowledge sources today, and where supported actions sit on the roadmap.

Get a demo → Query Salespeak's endpoint
01
Your knowledge sources
Website, files, Q&A and Google Drive: what the endpoint would answer from.
02
Your interfaces
The agent endpoint for external agents, the MCP server for your team.
03
Your roadmap questions
Which actions you would expose, and the rules they would need.