# Vendor AI Agent Logs: Who Owns Them and What to Export

> Running Agentforce, Fin, Zendesk AI agents or another vendor's agent? Who controls the session logs, what you can export and what stays hidden.

Canonical: https://origindatapartners.com/guides/vendor-ai-agent-logs-ownership
By: Origin Data Partners
Status: Published
Published: 2026-11-15
Updated: 2026-11-15
Sources checked: 2026-11-15

## Quick answer

Vendor AI agent logs sit in the vendor's platform under your agreement with it. Your company usually controls the conversations and session records stored for it, and an admin can export them: Agentforce through its Session Tracing Data Model, Zendesk AI agents through a data export API, Intercom through its conversations API. What you rarely get is the model, its reasoning or the harness. Your customers' words are in every log.

## Key points

- Three sets of terms apply: the vendor's agreement, your customer terms and the privacy notice.
- Admins can export conversations and session events; model internals usually stay with the vendor.
- The vendor's own resolution label is a start, not a verified outcome.
- Ask the vendor in writing when the terms are unclear, before you describe the data.

## What do vendor AI agent platforms record?

Each turn of each conversation, the actions the agent took and how the conversation ended. The depth varies by vendor.

Salesforce says Agentforce's Session Tracing Data Model "structures every logged event in a session down to a single conversation turn, capturing user input, planner decisions, prompt and action flows, gateway inputs and outputs, errors and final output" ([Salesforce](https://www.salesforce.com/blog/agentforce-observability/)), and that "Agentforce Observability is now unmetered for every customer across all Agentforce SKUs". Intercom's conversation record shows whether the AI agent took part: it "Indicates whether the AI agent participated in the conversation." ([Intercom API reference](https://developers.intercom.com/docs/references/rest-api/api.intercom.io/conversations/retrieveconversation)). Zendesk's AI agent export carries conversation status and its own resolution field, such as "automated_resolution" : "resolved" ([Zendesk developer docs](https://developer.zendesk.com/documentation/ai-agents/getting-started/data-export)).


## What can you usually export from a vendor AI agent?

Conversations and their metadata, by an admin, through the vendor's API or data platform. Each vendor has its own route and limits.

| Vendor agent | Export route (from the vendor's docs) | Limits to know |
| --- | --- | --- |
| Zendesk AI agents | "The Data Export API lets you securely extract conversation data and metadata from your AI agents." ([Zendesk](https://developer.zendesk.com/documentation/ai-agents/getting-started/data-export)) | "exported data only covers conversations that ended on the given date"; "Only Zendesk admins can view and copy the organization id." |
| Intercom Fin | Read each conversation: the API "will return a single Conversation model with all its conversation parts." ([Intercom](https://developers.intercom.com/docs/references/rest-api/api.intercom.io/conversations/retrieveconversation)) | Without paging, "the response returns the 500 most recent conversation parts". The content data export "does not export raw message or conversation content." ([Intercom](https://developers.intercom.com/docs/references/rest-api/api.intercom.io/data-export)) |
| Salesforce Agentforce | Session Tracing Data Model tables in Data 360 ([Salesforce](https://www.salesforce.com/blog/agentforce-observability/)) | Queried inside your org by an admin; retention depends on your edition |
| Fin Agent API (Fin driven from your own app) | Status events such as resolved or escalated: "Fired when Fin's status changes (awaiting_user_reply, escalated, resolved, complete)" ([Intercom](https://developers.intercom.com/docs/references/rest-api/api.intercom.io/fin-agent)) | Your app receives events; keep your own copy as they arrive |


## What can't you see in a vendor agent's logs?

Usually the model, its full reasoning and the harness. The vendor chose the model, may route between several and treats its prompts and orchestration as its own. You see what the agent said and did, not always why.

That limits what the logs can be used for. A run where the model and [harness](https://origindatapartners.com/glossary/agent-harness) are unknown cannot be compared with another or replayed in a different system. Write down what you do know: the vendor, the product version or date range, and any model the vendor names in its documentation or your contract.


## Who owns the conversation and action logs of a vendor AI agent?

Usually your company controls the data stored for it, under the vendor's agreement, and your customers' content inside it is governed by your own terms with them. The vendor typically has its own rights to operate and improve the service, set out in the same agreement.

So three documents decide what you may do: the vendor agreement (data ownership, export, use restrictions), your customer terms and data processing agreements, and your privacy notice. This page is general information, not legal advice. Read each one, and have counsel confirm before any log leaves the company.


## How should you read your vendor agreement?

Look for five clauses, and write down what each says before you ask anyone else.

- Customer data: who owns data you and your customers put in, and what the vendor may do with it.
- Output: who owns what the agent produces, and any restriction on using it.
- Export and deletion: what you can take out, in what format, and for how long after the contract ends.
- Confidentiality: whether the vendor's prompts, configuration or documentation are its confidential information.
- Use restrictions: any clause on using the service or its output to build other products or models.


## Worked example: describing Zendesk AI agent logs

A fictional retailer runs Zendesk AI agents on chat and email. An admin pulls a month of daily exports and counts 41,000 conversations, of which 6,200 needed more than three actions, such as order lookups or refunds.

Its description reads: vendor and product, date range, number of conversations, share escalated to people, the vendor's resolution field, and a note that the model and harness are the vendor's and not visible. It adds that customer names, emails and order numbers would be removed before any sample is shown. No logs leave the company to write it.


## Common mistakes with vendor AI agent logs

Each of these is easy to avoid if you check before you export.

- Treating the vendor's "resolved" flag as proof the customer's problem was solved. Join it to reopen and refund records.
- Using a content or analytics export and assuming it holds the conversations.
- Pulling only recent parts of long conversations because of API page limits.
- Describing the logs before reading the vendor agreement.
- Forgetting that every log holds customer content. Redact before any sample; see the [redaction checklist](https://origindatapartners.com/guides/agent-trace-redaction-checklist).


## Vendor agent logs can also be a data asset

Logs from a vendor agent are worth describing when they are long and end in a checkable result: a refund issued, an order changed, a case escalated and solved. Here is what buyers Origin works with asked for in November 2026: real production runs with the model and harness stated. A vendor's logs may not show which model or harness ran, so say plainly what is missing, and read the vendor's terms first. See [non-coding agent traces](https://origindatapartners.com/guides/license-non-coding-agent-traces), [exporting Zendesk tickets](https://origindatapartners.com/guides/export-zendesk-tickets) and [licensing agent traces](https://origindatapartners.com/guides/license-ai-agent-traces).

We help you license your data. Origin Data Partners is an independent data partner: we tell you what your records could be worth and prepare the records, the terms and the license with you. The AI company that licenses them pays your company; a buyer may pay Origin when a deal closes.

Does your company hold records like these? [See what your data is worth](https://origindatapartners.com/check-fit).


## Questions and answers

### Does our AI agent vendor own our conversation logs?

Usually not. Most agreements leave customer data with the customer and give the vendor rights to run and improve the service. Your agreement decides, so read its data, output and use clauses. This is general information, not legal advice.

### Can we state the vendor's model name when we describe the logs?

Only what the vendor has told you in its documentation or your contract, and only if your agreement does not treat it as confidential. Otherwise say the model is the vendor's and not visible to you.

### Are vendor deflection or resolution metrics an outcome label?

They are a first signal, not a verified outcome. Join each conversation to what happened next: a reopened ticket, a refund, a repeat contact within a set window.

### Can we export historical sessions from a vendor AI agent?

Often, within the vendor's retention. Zendesk AI agent exports run one day at a time; Intercom conversations are read through its API with paging. Check how far back each vendor keeps data before you plan.

### Should we ask our AI agent vendor before describing the data to anyone?

Yes, if the agreement is unclear on data use or output. Ask in writing what you may export and use. A written answer avoids a dispute later.

### Do Intercom Fin or Zendesk AI agent logs include tool calls?

They include the actions the agent took as conversation events or parts, and the outcome the vendor recorded. They usually do not include the model's prompts or reasoning, so they are thinner than a trace from an agent you run yourself.

## Sources

- [Salesforce blog: Agentforce Observability and the Session Tracing Data Model](https://www.salesforce.com/blog/agentforce-observability/)
- [Zendesk developer docs: exporting AI agent conversation data](https://developer.zendesk.com/documentation/ai-agents/getting-started/data-export)
- [Intercom API reference: retrieve a conversation](https://developers.intercom.com/docs/references/rest-api/api.intercom.io/conversations/retrieveconversation)
- [Intercom API reference: data export (content data)](https://developers.intercom.com/docs/references/rest-api/api.intercom.io/data-export)
- [Intercom API reference: Fin Agent API](https://developers.intercom.com/docs/references/rest-api/api.intercom.io/fin-agent)

## Related reading

- [How to Export Zendesk Tickets with Comments and History](https://origindatapartners.com/guides/export-zendesk-tickets)
- [Agent Traces: Licensing Production AI Agent Runs (2026)](https://origindatapartners.com/guides/license-ai-agent-traces)
- [PII Redaction for Agent Traces and LLM Logs: A Checklist](https://origindatapartners.com/guides/agent-trace-redaction-checklist)
- [License workflow data for AI agents: a company owner’s guide](https://origindatapartners.com/guides/license-workflow-data-for-ai-agents)
- [Export Agent Runs: Agno, AgentCore, Agentforce, OpenAI SDK](https://origindatapartners.com/guides/export-agent-runs-from-agent-platforms)
- [/glossary/agent-harness](https://origindatapartners.com/glossary/agent-harness)
- [Non-Coding Agent Traces: What Buyers Ask For (2026)](https://origindatapartners.com/guides/license-non-coding-agent-traces)
- [Can a Hosting or CDN Provider License Incident Data to AI?](https://origindatapartners.com/guides/hosting-cdn-provider-incident-data-ai)
