A publishing system built around one constraint.
Qalam is a LinkedIn publishing system with voice memory. The constraint is that output has to sound like the person whose name is on it. Everything else in the product exists to serve that, or it does not get built.
Why voice memory is the whole problem
Generating a clean, competent LinkedIn post is now close to free. That is precisely why a clean, competent post no longer signals anything. When the cost of polish drops to zero, the remaining signal is whether the writing sounds like a specific person with a specific view.
That is a storage problem more than a generation problem. Qalam holds the material the answer depends on. You save writing examples you are willing to be judged by, along with your role, industry, and the audience you write for. Those samples are analysed into tone, sentence length, vocabulary, and structural characteristics. When you draft, the most relevant examples are retrieved and used as the reference for that specific piece.
The profile changes when you change it, not silently in the background. That is a deliberate design choice: a voice that quietly retrains on every edit is a voice you cannot predict three months from now.
What the product refuses to automate
No autonomous posting
Qalam does not post, comment, react, or follow on its own. Publishing and scheduling are actions a person takes, and a scheduled post goes out at the time that person set.
No engagement automation
There are no pods, no automated comment loops, and no artificial activity. The comment assistant drafts options that a person reads, chooses, and submits themselves.
No silent voice drift
The voice profile changes when you change it. Qalam does not quietly retrain on your edits or post performance, which is what keeps the output predictable over months rather than weeks.
No unreviewed output
AI drafting is grounded in the evidence and examples held in the workspace, and a person still decides what gets sent or scheduled. Qalam does not publish anything you have not acted on, because no language model can be guaranteed never to overstate. Workspaces that want a second pair of eyes can add a reviewer approval step on top.
Voice before volume
A feature that helps someone publish more while sounding less like themselves does not ship. Output quantity is easy now. Recognisability is the scarce thing.
Say what is manual
If a workflow requires a person, we call it manual. If a score is a heuristic, we label it a heuristic. If a capability is not built, the page says it is not built.
Continuity through context
Evidence, saved writing examples, drafts, versions, and outcomes stay connected across the workspace, without pretending that every action silently retrains a model.
Verify the claims yourself
A page that describes a product is weaker evidence than the product. These surfaces are public and current.
- Interactive demoThe real interface running on labelled sample data.
- ChangelogWhat shipped, and when.
- Plans and limitsExact monthly allowances per plan, not the word unlimited.
- Scoring methodologyHow the diagnostics are calculated.
- System statusCurrent health of the running service.
- Privacy policyHow workspace data is handled and stored.
Running LinkedIn for more than one account?
Agencies and content teams give each client an isolated workspace with its own voice profile, archive, reviewer, and schedule.