Teams running AI agents in production
Sell agent logs to AI, after checking the platforms your traces pass through
- Subject
- Ai agent software data terms
- Date read
- Basis
- 19 vendors' own documents, 20 read
An agent leaves a trail in every tool it touches: the observability platform that stores its traces, the agent builder that runs it and the model API that answers it. We read 20 documents from 19 of these vendors, newest on 2026-10-28: 6 write that they will not train on what you send them.
Source: Ai agent software data terms: every row, with its source
Tell us about the agent logs you keep
What happens next
-
Say what your agents do, which platforms hold their traces and how long you have kept them.
-
We check those platforms' clauses and the logging rules that apply to your agents.
-
You hear back on which deal suits your traces: licensing them while the company operates; selling them as a closing company's records, which its directors can do only until a liquidator or administrator is appointed (a solvent members' liquidation included) and the office-holder then does, with one catch: a sale made after a winding-up petition is void if the company is then wound up, unless the court allows it (in a US case a bankruptcy trustee sells); or a route set with the adviser who referred you.
No agent platform, model provider or buyer has paid to be on this site or knows its pages were read, and no link here is an affiliate link. Each answer is our reading of a published page and is not legal advice on your own agreements. What we charge: if a company licenses its agent logs to a buyer we introduced, our fee is 12.5% of the deal when we act as its exclusive agent and about 10% when the logs go into a bundle or the pool, agreed in writing before anything is signed; anything a buyer pays us is credited to the company.
What each agent platform's terms say about your traces
Last updated
One row per agent builder, observability tool or model API, read from its terms, data controls or privacy page and linked to the page read. Not stated means that document did not answer the question.
- 6 of the 19 agent platforms and model APIs we read write that they will not train on customer data; newest read 2026-10-28.(Ai agent software data terms: every row, with its source)
| Agent platform or API | Who owns what you send | Keeps aggregated or de-identified data | Training on your traces | Export window at exit | Source |
|---|---|---|---|---|---|
| Ada | the business | reserved | not stated | not stated | Ada Customer Terms of Use |
| Amazon Bedrock | not stated | not stated | not stated | not stated | Amazon Bedrock, data protection |
| Anthropic (API) | not stated | not stated | will not train | not stated | Is my data used for model training? |
| Arize AI | the business | reserved | not stated | not stated | Arize AI Terms of Service |
| Braintrust | not stated | not stated | not stated | not stated | Braintrust Terms of Service |
| Cognigy | the business | reserved | not stated | not stated | Cognigy terms |
| CrewAI | not stated | not stated | not stated | not stated | CrewAI Terms of Use |
| Cursor | not stated | reserved | will not train | not stated | Cursor Privacy Policy |
| Galileo | not stated | reserved | may train | not stated | Galileo Terms of Service |
| Helicone | not stated | not stated | not stated | not stated | Helicone Terms of Service |
| Kore.ai | the business | not stated | not stated | not stated | Kore.ai Terms of Service |
| LangChain (LangSmith) | the business | reserved | not stated | 30 days | LangChain Terms of Service |
| Lindy | not stated | not stated | will not train | not stated | Lindy Terms of Service |
| Microsoft (Azure AI Foundry) | not stated | not stated | will not train | not stated | Azure AI Foundry, data, privacy and security |
| OpenAI (API) | the business | not stated | will not train | not stated | Data controls in the OpenAI platform |
| Replit | not stated | not stated | not stated | not stated | Replit Terms of Service |
| Vellum | the business | not stated | not stated | not stated | Vellum Terms of Service |
| Weights & Biases (Weave) | not stated | not stated | not stated | not stated | Weights & Biases master terms |
| Writer | not stated | reserved | will not train | not stated | Writer Terms of Use |
Method and citation
Each vendor's own published terms were read through estate-fetch and stored whole (sha256 on every row). For every document, four questions were answered from the vendor's own words: who owns the business's data, vendor keeps aggregated or anonymised use, training AI on the business's data, export at exit. Every stated answer carries the sentence it rests on, checked against the stored text; 'not stated' means the document read does not address the question. A vendor with several documents reads 'varies' where they disagree. 19 vendors, 20 documents.
“What each agent platform's terms say about your traces”, Sell Agent Logs to AI, https://sellagentlogstoai.com/ (updated 2026-10-28).
Read next
Every figure sourced and dated.
- LangSmith data retention, and getting traces out of other tools
LangSmith data retention and export after you leave, beside what Arize, Braintrust, Galileo, Helicone and other agent tools say about the traces they hold.
- Agent trajectory dataset: who takes one, quoted from their pages
Who takes an agent trajectory dataset from a company: trace licensers, RL environment builders, data partners and a model lab's data-sharing programme, read.
- AI agent audit logs: what the EU AI Act says to record and keep
AI agent audit logs under the EU AI Act: automatic logging and a six-month minimum, applying from December 2027 or August 2028, and California's automated-decision rules.
- LLM agent traces: what the model APIs say about training on them
What the model APIs and agent builders your LLM agent traces pass through say about training on them, from OpenAI and Anthropic to Writer, Lindy and Cursor.
- Can I sell my AI agent's logs? Yes, the runs you own, with three checks
Can I sell my AI agent's logs? Who buys agent trajectories and what one pays, what UK GDPR says about the people inside a log, and the model providers' usage rules.
The questions we get
Can a company sell the logs its AI agents produce?
It can license what it owns and is free to share. On this board 7 of 19 vendors say the data you send stays yours, but a trace also holds whatever your users and customers typed, so personal data has to come out before anyone licenses it.(Ai agent software data terms: every row, with its source)
Do model APIs train on our agents' prompts and completions?
The large ones we read say not by default. OpenAI's data controls say data sent to its API is not used to train its models unless you opt in, and Anthropic says it will not use inputs or outputs from its commercial products to train its models by default.(Ai agent software data terms: every row, with its source)
Does an observability tool keep a copy of our traces?
Some keep a version. Arize says it owns de-identified, aggregated or anonymised data derived from customer data, LangChain may disclose operational metadata externally only once it is aggregated or anonymised, and Galileo takes a right to use customer data to train its algorithms internally.(Ai agent software data terms: every row, with its source)
How long can I get my traces back after leaving LangSmith?
LangChain's terms let a customer export its data during the subscription or within thirty days after it ends, on written request and subject to the retention period it configured. Most other platforms we read state no export window, so pull the traces you want before you leave.(Ai agent software data terms: every row, with its source)
Does the EU AI Act make us keep agent logs?
For high-risk AI systems, from the dates the amended Act now sets: Article 19 makes providers keep the automatically generated logs for at least six months, and Article 26 asks the same of deployers, to the extent the logs are under their control and unless other law says otherwise. Regulation (EU) 2026/1744 moved those duties to 2 December 2027 for systems listed in Annex III and 2 August 2028 for those under Annex I.(Ai agent software data terms: every row, with its source)
Who takes agent traces and trajectories?
Magma says it packages an agent's interaction data for AI labs and pays a revenue share as it licenses; Ooak Data turns real company data into reinforcement learning environments for agents; OpenAI offers some organisations complimentary daily tokens for traffic they share with it.(Ai agent software data terms: every row, with its source)
Sources
This site's own datasets, every figure quoted from its source and dated:
- Who takes AI agent traces and trajectories: 6 of 14 organisations whose pages we read name agent traces, trajectories or workflows among what they take, read 28 October 2026.
- The law on keeping and passing on AI agent logs: 8 provisions of EU, UK and US state law and guidance on recording, keeping and disclosing what AI systems do, each quoted from the authority's own page, read 28 October 2026.
- Can i sell my AI agents logs (5 figures): 5 provisions of UK GDPR and the Information Commissioner's guidance on reusing personal data that agent logs hold, each quoted from legislation.gov.uk or ico.org.uk, read 28 October 2026.
- What model providers' usage rules say about training on…: 3 model providers' own usage documents restrict using what their models produce to train another AI model, each rule quoted from the provider's page, read 28 October 2026.
- Ai agent software data terms: 7 of 19 ai agent software vendors whose terms we read reserve the right to use their customers' data in aggregated or anonymised form, read 28 October 2026.