What is kept, what leaves, what deleting removes

Three questions a risk review asks first: how long is this kept, who else sees it, and what happens when we delete it. The answers below describe what the platform does, not a policy it aspires to - each is current at the time of writing.

Retention

Data	Kept
Conversation transcripts, attachments	Until deleted, unless your workspace sets a window (below). The platform itself never prunes them.
Tool-call records (each action's input and output)	90 days by default, then pruned daily. A workspace can shorten it to as little as 30 days.
Cost, per day	Kept: each closed day is rolled up before its raw records are pruned, and the prune waits for the rollup.
Per-turn census behind Health (tool names, counts, tokens, durations)	Kept - facts, never bodies - unless your workspace sets a window (below).
Run history (finished agent and flow runs)	Until deleted, unless your workspace sets a window (below).
Knowledge articles that were never published	Until deleted, unless your workspace sets a window (below). Published articles keep their history.
Audit trail (sign-ins, approvals, credential changes, support access, membership changes)	Permanent. An audit row can never be edited or deleted, by anyone.
Files uploaded but never sent	24 hours, together with the empty draft conversation created to hold them.
Data exports	7 days after they are ready, then the file is deleted.
Caller-supplied credentials	At most one hour, deleted when the run they were supplied to ends.

Your workspace's own windows

Owners and Admins can set how long the workspace keeps its records, under Settings → Data → Retention. There is a window for each kind of record:

• Conversations are deleted whole, with their attachments, once their last message is older than the window.
• Attachments covers files in conversations you keep.
• Memories covers personal and agent memories.
• Ratings.
• Tool-call records. This window can only be shorter than the platform's.
• Run history covers finished agent and flow runs. A run stays until its tool-call records are gone, so its cost history is never lost early. Runs still in progress are never deleted.
• Agent health records covers the per-reply facts behind each agent's Health page.
• Unpublished knowledge covers knowledge drafts and failed uploads that were never published, counted from their last edit. Published articles and their version history are not affected.

Each window is 30 to 3,650 days, or unset, which keeps that kind of record until someone deletes it. A value below 30 is refused rather than rounded up. Preview impact counts what the next nightly run would delete, using the same rules the run applies. The run happens once a day, and every change to the windows is recorded in the audit trail. The audit trail itself is never pruned.

What the prune changes on screen is described in Watch and inspect tool calls.

Deleting a conversation

A visible conversation can be deleted by its creator/recorded owner or a domain Owner through the API (DELETE /sessions/:id). Domain membership is still required for domain sessions, including for their creator. Organization Owners/Admins have no general deletion bypass, and personal chats remain private to their creator. Historical sessions from the retired Apps feature retain organization Owner/Admin control. See Roles and permissions. It removes the transcript, every recorded action, and every attachment together with its stored bytes; a flow run started from it keeps its own record, detached.

Two things deliberately survive, and an erasure obligation has to reach them separately:

• A test saved from that conversation keeps its own copy of the transcript - that copy is what lets the test replay months later. Remove the test from the agent's Evals as well.
• Memory the agent saved during the conversation stays in the agent's memory; it is attributed to the conversation but not owned by it. Delete the case from the agent's Memory tab.

The audit trail is untouched by any deletion.

Deleting other things

• A knowledge file on an agent removes its chunks and its bytes; retrieval re-indexes on its own.
• An organization article that was ever published cannot be deleted - its versions are history - but it can be archived. Drafts can be deleted.
• A memory case removes the whole revision chain for that key. A learned rule is removed outright.
• A rating you gave a reply is yours alone to clear; a session's outcome can be withdrawn.
• A person can be deactivated, which keeps their conversations, messages and audit rows, or erased, which deletes what only they used and anonymises their words everywhere else. See Data-subject requests.

What leaves your workspace

On a model call, the provider receives what the turn's context meter shows: the base prompt, the agent's Principles and Procedure, the knowledge passages injected or retrieved for that turn, the rolling summary, the trimmed history, image attachments as images, the definitions of the tools presented, and every tool result. It runs on whichever key the provider is set to under Settings → AI settings → AI providers: your own key, under your agreement with that provider, when the provider is on Own key; Yekar.AI's key, drawn from your prepaid Credit balance, when it is on Managed. There is never a silent switch between the two - a provider that cannot serve the call fails as itself. See Your keys, your model bill.

What the model provider keeps

Wherever a provider's API lets a request say "do not keep this", Yekar.AI says it on every model call - agent replies, flow steps, and the background work Yekar.AI does for you such as conversation titles, summaries, procedure structuring, search-query rewriting, document reading and decision steps. Beyond that, what a provider keeps is set by its own terms for the account whose key ran the call: your agreement with the provider on Own key, Yekar.AI's on Managed. If you need a provider to keep nothing at all (sometimes called zero data retention), that is an arrangement you make with the provider for your own account, and you then use it on Own key.

Provider	What Yekar.AI asks on each call	What that means
OpenAI	Do not store the response (store: false)	OpenAI does not keep the reply for later retrieval or for its stored-completions features. Without it, OpenAI's newer API keeps each response for at least 30 days. Any retention OpenAI applies for abuse monitoring is governed by its terms, not by the request.
OpenRouter	Use only upstream providers that do not collect the data (data_collection: "deny")	OpenRouter routes the call only to providers that do not store or train on it. A model with no such provider behind it on OpenRouter can fail to run there; pick another model or provider if that happens. Applied as OpenRouter documents it; not yet confirmed against a live OpenRouter request.
Anthropic, Google (Gemini), Amazon Bedrock	Nothing - these APIs have no per-request setting	Retention follows the provider's terms for the account. Zero retention, where offered, is agreed with the provider for an account rather than set on a request.
Every other provider (Groq, xAI, Mistral, DeepSeek, Vercel AI Gateway and the rest of the list in AI providers)	Nothing - none documents a per-request setting Yekar.AI could confirm	Retention follows that provider's terms for the account. Yekar.AI does not send a setting it cannot confirm the provider accepts, because an unknown setting can make every call fail.
TypeSafe AI (decision models)	Nothing - its API has no per-request setting	Retention follows TypeSafe AI's terms for the account.

Four more services receive text from your workspace and have no "do not keep" setting to send:

• Knowledge and memory embeddings run on Yekar.AI's OpenAI account. OpenAI's embeddings API has no store setting, so OpenAI's terms for that account apply.
• Web search sends the search query to Serper on Yekar.AI's account. It is on by default in every workspace where Yekar.AI offers it, and every search run in your workspace - by an agent, a flow's web search step or a conversation - goes this way. Serper's API has no retention setting.
• Reading a page through Jina AI, where your workspace has that tool, sends the page address or search query to Jina AI. Its API has no retention setting.
• Media generation, where your organization has it, sends the generation prompt to fal.ai on Yekar.AI's account; the generated file comes back and is stored as an attachment. Its API has no retention setting.

Yekar.AI does not restrict which providers you can use on these grounds; the table is here so you can choose.

Knowledge and memory search processes document and article excerpts, saved memories, tool descriptions, and search queries to find relevant matches. Usage records contain counts and character totals rather than the text itself.

Web search uses your query to find titles, links, and snippets from public pages. Browser reads the public URL you provide and returns its page text. Integration tools send the information needed to perform the requested action to the connected service. See Platform security.

Where the bytes live

Records and transcripts live in Yekar.AI's own storage. Files - attachments, uploaded documents, generated media - live in object storage under an opaque key; downloads are short-lived signed links. Credentials are encrypted with a per-organization key that is itself wrapped by a master key, decrypted only at the moment a call needs them, and never readable back through the API: every secret shows as its last four characters and nothing more.

Export

A person's own data - profile, conversations, uploads, memories, ratings and runs - exports as a ZIP of JSON files from Settings → My data, and an Owner or Admin can prepare the same export for any member. See Data-subject requests.

An organization's knowledge - bases, articles with their draft and published bodies, visibility and modules - exports as one JSON snapshot through the API (GET /domains/:domainId/knowledge-export) and imports through Import in the knowledge library's actions menu, which creates new copies in the domain you choose. Transcripts are read through the API; the organization audit log is read and exported under Settings → Audit log or through GET /org/audit-events and GET /org/audit-events/export.

Product analytics

Product analytics runs only for people who choose to share it, and session recording additionally only in organizations whose Owner or Admin turned it on. See Privacy and product analytics.

Support access

When Yekar.AI's staff act inside your workspace, it is under a recorded grant with a mode, a stated reason and clocks, visible to you under Settings → Support sessions, where you can also turn support access off or require that one of your admins approves each session first. A summary email is sent when it ends unless a platform admin with full write impersonation access turned off Inform user for that session. See Platform security.