Cody

What Cody is

Cody is a business-conversation AI platform from the CodeSM group. Its current product family includes Cody Assist for knowledge and support, Cody Convert for proactive website conversations, Cody Clear for controlled responses, and Cody Live for avatar-based experiences. The official business-product site is meetcody.ai.

Assist is the most direct starting point for a team that wants questions answered from its own information. Users supply company content, configure an assistant and share it with employees or customers. The other products address different interaction styles and control requirements; they should not be assumed to come with every Assist account.

Cody's site makes strong claims about grounded answers and citations. Those describe its intended behavior, not an independent guarantee that every answer is correct. Evaluate the supplied content, retrieval and actual responses. A useful business assistant needs a source collection that represents the current policy as well as an interface that can retrieve it.

Cody Assist: a knowledge-based assistant

The Assist product page describes bringing documents, websites and other business content into a knowledge workspace. It supports focused knowledge folders and assistants with different roles, instructions and response styles. The purpose is to answer relevant questions without requiring the user to search through the source collection manually.

Choose the collection around the job. An employee-policy assistant needs the approved handbook, benefits rules and relevant process documents. A customer-support assistant needs public product instructions and support information. Mixing these collections indiscriminately can create confusion about which answer belongs to which audience.

Start with a small coherent set. Remove obsolete versions and identify the owner of each policy. If two documents conflict, resolve the conflict before treating the assistant as a source of guidance. Retrieval can locate information but cannot reliably determine which undocumented exception the organization intended to preserve.

Configuring specialized assistants

Assist's documented setup includes defining a role, instructions and guardrails for a particular job. That can make a knowledge assistant more useful than a single general conversation shared across unrelated departments. Configuration should explain the audience, scope and response expectations, including how to handle a question the sources cannot answer.

For an IT-support assistant, specify the supported applications and the level of help it may provide. A password-reset question might call for a link to the approved process. A request for another person's access should follow a different path. Use instructions that reflect the organization's actual workflow rather than asking the assistant to solve every issue independently.

Test whether the configured assistant stays within that role. Ask an ordinary question, an ambiguous one and one outside its collection. Check the response's usefulness and the source it selects. Clear configuration helps shape behavior, but the test should assess the actual result rather than assuming the instruction is perfectly enforced.

Citations and answer review

Cody presents source attachment as a core Assist feature. A citation lets a user inspect the document behind an answer, which is especially useful when a policy has conditions or exceptions. It also gives the content owner a way to investigate a questionable response.

Open the source and compare the wording with the claim. A document can discuss a subject without supporting the precise answer. For example, a return-policy page may describe one product type while the question concerns another. Check whether the answer carries over the relevant scope, dates and exclusions.

Use failures to distinguish content and retrieval problems. If the correct policy is missing, add or update the approved information. If the policy exists but the answer chooses an older source, examine organization and configuration. If the answer misinterprets the cited passage, keep that example for evaluation. Citations make review possible; they do not eliminate the review.

Sharing Assist with teams and customers

Assist documents shareable links, inline embeds and popup website widgets, with integrations and API access for additional uses. The current product pages also describe use in collaboration channels such as Slack. The right delivery route depends on where questions already arise and how the audience is authenticated.

An internal assistant and a public website widget have different information boundaries. Confirm which knowledge the shared assistant can retrieve, who can access it, and what conversation information is recorded. Do not assume that a public embed automatically reproduces employee permissions. Test with accounts or sessions representing the intended users.

Keep the interaction easy to interpret. The user should know which organization and source collection the assistant represents, and how to contact a person when necessary. For a support widget, check its behavior on the actual page and device. A working embed is only the interface step; knowledge quality and support ownership remain part of deployment.

A concrete workflow: an employee travel-policy assistant

Imagine an operations team answering repeated questions about business travel. Gather the approved reimbursement policy, booking instructions and expense-submission guide. Remove superseded documents and clarify who owns exceptions. Organize the collection so the assistant's intended topic is clear.

Configure Assist to answer travel-policy questions from those sources and explain when a question requires a manager or finance contact. Test a standard hotel-expense question, a request outside the stated allowance and a question about a policy that is not documented. Inspect whether the answer includes the relevant conditions and a useful citation.

Share the assistant through the confirmed internal route. Give employees a way to report an unclear answer and assign someone to review those examples. When the reimbursement rules change, update the source collection and retest the questions affected by the change. Do not rely on a general claim of website sync for a document that is maintained manually.

This workflow illustrates a repeatable knowledge service rather than an autonomous policy decision maker. The assistant can help employees find the documented rule. Managers and finance staff remain responsible for exceptions and changes. Its usefulness depends on current content, scoped access and a manageable path for correcting answers.

Cody Convert and early-access website engagement

The Convert page explicitly labels the product early access. It describes using visitor context, such as source and viewed pages, to initiate a conversation and guide a visitor toward a next step using approved content. Its stated purpose differs from an Assist widget that waits for a knowledge question.

For a team evaluating this route, define the intended interaction. A visitor reading pricing may need a concise explanation of the plans, while a returning prospect may need a demo request. Decide which information the agent may collect and which actions it may take. An early-access description is not a promise of universal production availability.

Assess engagement with a clear baseline and the actual website audience. More conversations do not automatically mean better outcomes. Check whether the interaction helps users complete their task, whether recommendations match approved information, and how a person can continue the conversation. Vendor conversion claims should not be treated as measured results for the evaluating site.

Cody Clear and controlled response design

The Clear page describes selecting approved responses, semantic selection from an approved set, and optional generation behind configured gates. It also describes versioned controls and traceability. This is a distinct design for organizations that need more control over wording than an ordinary generative knowledge assistant.

Evaluate the approved-answer process first. Who writes the answer, who authorizes it, and how does a changed policy replace the prior version? Test a straightforward match, a paraphrased question and a request that should remain unanswered. The selected answer must fit the user's question as well as being approved in isolation.

Clear is marketed for controlled and regulated contexts, but that label does not establish legal compliance or fitness for every such use. Verify the actual deployment, review process, data handling and records with the responsible team. The controlled-response architecture and access agreement need their own assessment; Clear should not be assumed to be included in Assist subscriptions.

Cody Live: face and voice

The Live product adds a branded visual character and spoken interaction to a business agent. The vendor describes website, physical-space and event uses, with actions such as presenting information, booking meetings or collecting a lead through configured integrations. The interface changes how people encounter the knowledge, not the need to define its scope.

An event assistant needs a different evaluation from a text widget. Test the environment's audio conditions, screen placement, conversation timing and fallback when someone cannot hear or understand the response. Confirm what contact details are collected and how the person is informed. A visually engaging character is useful only if the interaction serves the visitor's task.

Prepare a narrow content collection for the venue. An exhibition guide may answer opening hours and describe demonstrations, while routing purchasing questions to staff. Verify the specific actions and delivery agreement. Do not infer that every advertised event or avatar option is available through the same self-service setup as Assist.

The Cody API and integration work

The official API reference lists operations for bots, conversations, messages, documents, folders and uploads. These are relevant when a developer wants a custom interface or a content-management process around the Cody service. Review authentication and rate limits for the intended integration.

For a custom support interface, distinguish the application's user identity from the service credential used to call the API. Define which assistant and knowledge collection the request may reach. Check how conversation state is created and how errors are shown. An integration should preserve the audience boundaries established in the underlying setup.

The API documentation portal also offers MCP access to its documentation. That is an interface for reading the reference, not proof that the entire business agent is automatically installed as a runtime MCP tool. Use the documented endpoint and operation for the actual integration rather than inferring functionality from the portal's navigation.

Suitable users and choosing among the family

Cody is relevant to operations, HR, IT and support teams with a defined collection of business information. Assist is a candidate for employee and customer knowledge questions. Convert addresses proactive website engagement, Clear focuses on controlled wording, and Live adds an avatar experience. Match the product to the interaction before comparing plans.

Teams without maintained documentation should expect content work before a dependable launch. Start with common questions and the sources needed to answer them. A well-scoped assistant covering a small collection can be more useful than a broad agent whose information is contradictory or unclear.

Consider another workflow platform when the requirement primarily concerns complex action orchestration or a particular channel integration. Cody's knowledge approach is relevant when retrieval and conversation are the central job. The brand's wider family should not obscure the difference between supplying answers and operating business actions.

Pricing and access

The pricing page presents free starting access for Assist and paid plans with usage credits, team and content limits. Website synchronization, model options and conversation-history limits differ. Check the current plan for the intended collection and delivery route instead of assuming all advertised functions are in the free tier.

The other products use their own access or contact arrangements; Convert is labelled early access, while controlled and live deployments require their relevant product information. The enterprise page describes solution-design work for more specific deployments. A general Assist price should not be used as a quote for the whole family.

Estimate usage with typical questions and the expected audience. Include follow-up messages and periods of high demand, and identify the plan's content-update process. Assist has a free entry point and paid plans; confirm separate arrangements for specialized products.

Limitations and frequently asked questions

Cody depends on source quality, appropriate access and the selected interaction design. Grounding and citations make answers inspectable but do not guarantee correctness. Proactive engagement, controlled responses and avatars involve different configurations and access arrangements.

Who operates the business Cody platform?

The CodeSM group operates the business-conversation platform at meetcody.ai, including its Assist, Convert, Clear and Live products.

Can Cody answer from company documents?

Assist documents business-content ingestion, specialized assistants and answers with source references. Review the content and resulting answers.

Is Convert generally available?

Its current page labels it early access. Confirm access and operating terms for the intended website.

Does Clear guarantee compliance?

Compliance depends on the deployment and the responsible team's requirements. Assess the approved-response process, controls and applicable agreement.