What Iridium does with your data
Short answer: your LinkedIn password is never stored, other people's profile data is purged on a daily timer whether or not you run anything, and deleting your account kills every session immediately and stays reversible for 30 days. Some things are kept on purpose, and this page says which ones and why.
Credentials
Your LinkedIn password is never stored. There is a password-free connect path where you authenticate on a hosted page and Iridium ends up holding an opaque handle rather than a credential of yours.
| Credential | How it is handled |
|---|---|
| LinkedIn password | Never stored. A password-free hosted connect path exists. |
| Access token | Lives 15 minutes. |
| Refresh token | Rotates on every use, with theft detection. Reusing an old one revokes every session you have. |
| API key | Stored only as a SHA-256 hash. Shown to you once and never again. |
| Anything secret | Credentials, cookies and JWTs are never written into an audit payload. That is an invariant in the code, not a habit. |
The refresh token rule is worth spelling out. Tokens rotate, so an old one is supposed to be dead. If a dead one is presented, that is evidence somebody copied it, and the response is to revoke the whole family rather than just refuse the one request. Theft costs the thief the session and costs you a login.
Retention
Discovered profile and post data is purged after seven days by a job that runs every day at 07:00 UTC, platform wide. It runs whether or not you started a pipeline that day. Nothing about the timer depends on you remembering it.
Your in-app agent chat transcript is deletable on demand. You do not have to email anyone to get it.
Deletion
When you delete your account, four things happen in this order.
- The LinkedIn session is destroyed immediately.
- Every session on every device is revoked immediately.
- Your personal data is erased within 30 days.
- For those 30 days the deletion is reversible. Reactivate and your account comes back.
During the grace window, any API call returns a 403 that names your exact erasure date. You are told the day, not a policy paragraph.
Deleting Iridium never touches your LinkedIn account. Your profile, your connections and your messages on LinkedIn are unaffected.
What we keep on purpose
Account deletion does not delete everything, and saying otherwise would be a lie that is easy to check.
Audit, outreach, comment and pipeline logs are retained, and the user row survives. That record is the only thing that can answer the question of what was sent from this account and on what day. A product that sends messages on your behalf and then destroys the evidence of what it sent is worse for you, not better. So the log stays.
The activity log is append-only by design. Entries are written and not edited. We will not call it immutable or tamper-proof, because those words describe a cryptographic guarantee and this is a design rule.
The log is also built to hold as little of other people's data as possible. Audit rows deliberately do not copy third-party names, on the reasoning that doing so would make third-party personal data permanent.
Isolation
| Boundary | How it holds |
|---|---|
| Between users | Cross-user handle resolution is impossible by design. One account cannot resolve another account's handles. |
| Raw LinkedIn ids | The MCP server never returns them. |
| Activity log | Scoped to the authenticated user at every entry point, including the ones that are easy to forget. |
| Database | Postgres is published only on the private IP. It is not on the public interface at all. |
Analytics and legal pages
Product analytics runs cookieless. All text is masked. There is no session replay and no identify call, so events are not tied back to a named person. Global Privacy Control is honoured.
The terms and privacy pages carry no analytics at all. Reading the legal text should not be a tracked event.
We do not sell your personal information. Subprocessors are named individually in the Privacy Policy, each with what it is used for.
The terms you agreed to are pinned
When you accept the terms, a SHA-256 hash of the exact bytes served to you is recorded against your account. It proves what you agreed to, not merely that you clicked. If the terms change later, the record of your agreement still points at the version you actually read.
Safety rails, and why they are in the server
A limit written into a prompt is a suggestion. Every limit here is a check in the server, and it is identical whether the request comes from the web app or from an agent.
| Rail | Value | What happens at the edge |
|---|---|---|
| Comments per day | 25 | Deferred to the next morning |
| Outreach per day | 25 | Deferred to the next morning |
| Invites per rolling 7 days | 90 | Deferred to the next morning |
| Auto-approve | Does not exist | Every send is downstream of an explicit approve call |
| AI text approved unedited | Blocked for comments and posts | The server returns an error telling you to make it yours first |
| Multi-step sequences | Shipped disabled | Disabled on purpose, because they would bypass the quality check and the recipient cap |
The deferral behaviour matters more than the numbers. When a send would breach a cap it is not thrown away and it is not silently retried into the wall. It moves to the next morning. You lose a day, not the message.
If you are a recruiter
Candidate data has a shelf life and the shelf life runs on a timer. Discovered profile and post data is purged after seven days by the daily job, and that job runs whether or not you ran a search that day.
The permanent record stores what you did rather than a dossier on the candidate. The dedup log exists so you do not contact the same person twice, and it holds your action, not their details. Audit rows deliberately do not copy third-party names, because copying them would make somebody else's personal data permanent inside our system. You get a durable record of your own behaviour and the candidate does not get a permanent file.
If you run a brand
You cannot approve AI text unedited. For comments and for posts, the server returns an error telling you to edit it and make it yours before approving. There is no way to hold down approve on a queue of generated text and call it a content strategy.
The caps are server side, so nobody on your side can raise them by asking an agent nicely. And you own what you create. The text that goes out is text a person edited and approved.
If you are job hunting
Deleting Iridium never touches your LinkedIn account. Your profile, connections and messages stay exactly as they are.
Deletion is reversible for 30 days, so leaving is not a decision you have to make perfectly the first time.
Your agent chat history is yours to wipe, on demand, without asking us.
One thing we will not tell you: that any tool can make your job search invisible to your employer. Nothing controls what LinkedIn shows other people about your activity. What Iridium controls is its own behaviour. It sends nothing without your approval, and it can skip engaging with open to work content so your public activity does not read as a search.
What we are not going to claim
Security pages are usually a list of things a vendor wants to be true. These are the ones we are not going to say.
| Claim we do not make | Why |
|---|---|
| Encryption at rest | Nothing in the deployment documents it, so we are not going to put the phrase on a page. |
| SOC 2 or ISO 27001 | We have not been through either audit. |
| Penetration testing or a bug bounty | Neither exists yet. |
| A DPO or an EU representative | Neither is appointed. |
| All your data deleted on account deletion | The audit logs are kept by design. See the section above. |
| EU-only data residency | We do not commit to it. The Privacy Policy states where processing may happen. |
| Expired drafts are deleted | They stop appearing in your queue. That is a different sentence and we are going to use the accurate one. |
| Automated data export or age verification | Neither is built. |
| A team or multi-seat story | Iridium is single tenant, one plan, one LinkedIn account. |
A vendor who claims all of these and can evidence none of them has told you something about how they handle the rest.
Questions people ask next
Does Iridium store my LinkedIn password?
No. Your LinkedIn password is never stored. There is also a password-free connect path, where you authenticate on a hosted page and Iridium holds an opaque handle rather than a credential of yours.
How long do you keep other people's LinkedIn data?
Discovered profile and post data is purged after seven days by a job that runs every day at 07:00 UTC, platform wide. It runs whether or not you started a pipeline that day, so the clock is not tied to your activity. The permanent dedup log records what you did rather than the person's details, and audit rows deliberately do not copy third-party names.
What happens when I delete my Iridium account?
The LinkedIn session is destroyed immediately and every session on every device is revoked immediately. Your personal data is erased within 30 days, and for those 30 days the deletion is reversible through reactivate. If you hit the API during the window, the 403 you get back names your exact erasure date. Deleting Iridium never touches your LinkedIn account.
Is everything deleted when I delete my account?
No, and this is deliberate. Audit, outreach, comment and pipeline logs are retained, and the user row survives. That record is what lets anyone answer the question of what was sent from this account and when. We would rather tell you we keep it, and why, than claim a clean wipe that is not true.
What are the sending limits and can an agent talk its way past them?
25 comments a day, 25 outreach a day, and a rolling backstop of 90 invites per 7 days. They are enforced server side and are identical whether the request comes from the web app or from an agent. A limit written into a prompt is a suggestion. This one is a check in the server. When a send would breach a cap it is deferred to the next morning rather than failed.
Is there an auto-approve mode?
No, nowhere in the product. Every send is downstream of an explicit approve call that a person makes. For comments and posts you cannot approve the AI text unedited: the server returns an error telling you to make it yours first. Multi-step sequences ship disabled on purpose, because they would bypass both the quality check and the recipient cap.
How Iridium does it
Iridium is a LinkedIn MCP server. Your agent gets the inbox, drafting, commenting, publishing and the activity log as tools. Every outgoing action is a draft, and approving is a separate step that only a person takes.
It is $20 a month for one LinkedIn account, with a seven day free trial and no card. The full detail lives in the Privacy Policy and the Terms.
Read the policies, then try it
7-day free trial. No credit card. Cancel anytime.
Start free → How approval works