Does Ashby Remember Between Runs? The Profile Persists. The Cross-Candidate Work Does Not.
Does Ashby Remember Between Runs? The Profile Persists. The Cross-Candidate Work Does Not. Short answer: Ashby keeps one candidate profile per person, and that profile holds a great deal. Every role the person was considered for, every project they sit in, the notes and feedback in the profile feed, and the stage history for each job consideration all survive between runs. What does not survive is the work a scheduled run does across many profiles at once. The comparison, grouping, and shortlis
Does Ashby Remember Between Runs? The Profile Persists. The Cross-Candidate Work Does Not.
Short answer: Ashby keeps one candidate profile per person, and that profile holds a great deal. Every role the person was considered for, every project they sit in, the notes and feedback in the profile feed, and the stage history for each job consideration all survive between runs. What does not survive is the work a scheduled run does across many profiles at once. The comparison, grouping, and shortlist logic built during one run are gone when it ends, unless saved somewhere the next run can load.
That gap is easy to miss, because Ashby feels well remembered from the inside. Open a profile and the history is right there. Close a run that just reasoned over two hundred profiles, and the reasoning has nowhere to live.
What one candidate profile keeps in Ashby
Ashby takes the opposite approach to tools that split a person into one record per application. The candidate profile is where all information about a candidate is stored. A single profile lists every role the candidate has been considered for and every project they are part of, and a reviewer can jump between job considerations from inside that one profile to see the interview progress for each.
The summary on the profile holds the contact and identity layer: email address, phone number, social links, candidate custom fields, and additional uploaded files. Around it sit the working details for each job consideration: current activities such as interviews to be scheduled or drafted emails, the stages passed through, the hiring team, and the source that brought the person in.
The profile also carries reminders. A follow-up can be set for a chosen date, with a note attached, so a candidate who should be revisited does not depend on someone remembering to check back.
That is durable, person-level memory. It answers who this candidate is, where they stand in each pipeline, and what they submitted. It does not answer what a run concluded about this candidate while looking at the rest of the pool.
What the feed, duplicates, and rediscovery add
The second layer is the Feed: a list of all notes and feedback left on the candidate profile. Feedback submitted after interviews lands there, notes added by the team land there, and the record stays attached to the person rather than to one application.
Ashby also guards the profile layer against splitting. Duplicate detection matches on email addresses, name, and social links such as LinkedIn, GitHub, and Twitter. When an application arrives with an email that matches an existing profile, the application is added to that profile instead of creating a second one. When duplicates do slip through, profiles can be merged: the profile the merge starts from keeps its candidate ID, every job consideration from both profiles remains intact on the merged record, and the merge cannot be undone.
On top of that sits AI Talent Rediscovery. When a new role opens, rediscovery scans the existing database for people who applied before, interviewed, or received strong feedback, and surfaces them in buckets such as high fit and warm leads. Criteria defined once for a role can be reused, so the definition of a good match persists.
Ashby has added one more recall aid: an AI Candidate History Summary in the Chrome Extension. It pulls together applications, sequences, interviews and feedback, offers, notes, talent projects, emails, and messages into a cited summary of about 150 words or fewer, so past interaction can be read before new outreach starts.
All of this is real retention, and it helps a human enormously. Notice the shape, though. Every layer is organized around one candidate, or around criteria for one role. None of them was built to hold conclusions that belong to a run: that one source is quietly producing the strongest candidates for two teams, that five stalled profiles share one missing piece of feedback, or that a shortlist was ranked in a particular order for reasons recorded nowhere.
What a scheduled run loses overnight
Consider a routine that runs each morning before recruiters start. It sweeps new applications across open roles, checks for likely duplicates the automatic matching may have missed, groups candidates considered for several roles, flags feedback on the same person pointing in opposite directions, and prepares a priority order for the day.
During that sweep, the run builds knowledge that exists in no profile. It decides that two close-but-not-exact name matches are one person who should be reviewed once. It notices a pattern across sources. It traces several separate stalls back to one shared cause, such as a single stage where feedback is consistently late.
When the run ends, the profiles are still complete. The feeds still hold every note. The duplicate call, the source pattern, and the shared cause were never written to a shared history the next run can load. The next morning starts from the raw profiles again and rebuilds the same groups and priority order before any new work begins.
The cost is not lost data. The data stayed in Ashby. The cost is repeated reasoning, paid in full on every run, for work that was already done well once.
Give the runs one shared memory
Vilix AI is built for that missing layer: one memory held outside any single candidate profile, which every run can load before it starts and update before it ends.
It is cloud-hosted, so there is zero infrastructure to run or maintain. No separate store has to be provisioned and keyed to candidate IDs by hand just so an automation can keep its own working notes between runs.
Every connected tool reaches the same memory over MCP. A morning screening run beside Ashby, an assistant drafting outreach, and a reporting step summarizing pipeline movement can all read and write the same Vilix AI memory. The same memory follows the work everywhere, across tools and devices, instead of ending at the edge of one profile.
Vilix AI stores full conversation history, not just facts extracted from it. For hiring work, that difference is practical. A stored fact can say two profiles were grouped as one person. The full history shows what the run compared, which match it rejected, and why a case was left open, so the next run can check the reasoning instead of repeating it.
With that layer in place, the morning changes. The first run saves its groups, patterns, and open decisions to Vilix AI. The second run loads that saved work first, then spends its effort only on new applications and cases still waiting. Ashby keeps the profile and the feed. Vilix AI keeps what the runs learned across them.
Start free, keep control of the data
Vilix AI has a free plan forever, with no time limit forcing a choice. Paid plans start with a 7-day Pro trial with no credit card required, so a full hiring cycle can be tested before any payment.
Data stays under the team's control. Everything can be exported at any time, and everything can be deleted at any time.