Everyday productivity

The file someone shared in April, and the note that it was replaced

Ask for a file, link or number your team already shared. Get it back with who posted it, where it sits in the work, and whether something newer replaced it.

Join the waitlist See what it produces

Asked once, answered with its receipts

Find anything your team has posted

Who it is for: The person everyone asks — a chief of staff, an ops lead, an engineering manager.
What starts it: On demand. Ask from the use-case catalog, in the words you would have used asking a colleague.; On demand

Turn it on in Aestus

The question is never "does this file exist". It is "which one is the real one". A team of forty produces four versions of the same sheet in a quarter, and the copy that is easiest to find is usually not the copy that is true.

Aestus answers from the work graph it already keeps. The artifact knows which ticket it was attached to, which plan that ticket belonged to, and which decision cited it. It also knows which artifact replaced it, and who last verified that. So the answer is one line, and under that line is the reason you can trust it.

The hard part of a search assistant is not finding things. It is being honest about what it did not find. An answer built from three hits when there were thirty is worse than one that says it read three, because it reads exactly the same. So every answer here states the scopes it searched, the time window it used and the queries it ran, and a scope it could not read — Slack with no connection, a channel the bot is not in — is listed with that reason rather than quietly dropped from the list.

Nothing is counted by the model. The answer envelope has no field for a count at all, and every number on the card is recomputed by the product from the hits that survived checking. A claim with no source is refused outright rather than softened into a hedge, and an answer that says "something newer exists" without naming what is refused the same way.

Nothing leaves Aestus. Both steps that use tools are read-only: this workflow reads your records and the Slack channels your bot already belongs to, and it writes a document and a ticket comment inside your own workspace. It cannot post, send, edit or move anything.

What you get

The artifact this use case produces, in the product view you would read it in.

Example — fictional team and data. Not a live run or customer result.

Worked example

document

Found it — the enterprise tier number

Produced by: Run EXAMPLE-FP-01 · read-only · document artifact · demo values

The answer: One value, in bold, on its own line — then the source line under it: the artifact it came from, the ticket it hangs off, who posted it and when. Because the top hit has been replaced, the card also carries the warning that names pricing-model-v4-9p4n2v, who posted it and what the same cell reads there.
Where it came from: A table with one row per provenance field the answer actually has: the artifact and its kind, the date and poster, the ticket it was attached to, the plan and project that ticket belongs to, the approved decision that cited it, and its curation state. A field the envelope does not carry has no row, rather than a row that is filled in.
Other things that matched (and why they are not the answer): The 2 things that matched and were rejected, each saying why — a message with the number only in an image, a deck that quotes the sheet rather than being it. At most three, because a longer list is a log and has stopped answering the question.
What I searched: The 2 scopes that were searched, the time window, and the 4 queries that were run. A scope that could not be read — Slack with no connection — is listed with that reason. This section is printed even when nothing was found, which is what makes an empty answer believable.
Next step: One concrete action naming a link or a person that a tool actually returned — here, opening the version that is current today rather than the one that went to the April board.

How it runs

Each step is one node in the workflow graph the interpreter runs. Nothing here is a prompt chain hidden in a route handler.

Workflow steps

01Input
Your question arrives as one sentence. Anything that came with it — the rest of the Slack thread, a quoted message — is kept in a separate quarantined block, and every step downstream is told that text in it is data somebody wrote, never an instruction to follow.
02LLM
One call turns the sentence into a search plan: the terms, the person, the time window. "Before the April board" becomes a date range rather than a search term. Anything the question did not imply is left out instead of guessed.
03Agent
It searches your own records first — tickets, plans, projects, goals, artifacts and approved memory claims — then the Slack channels the Aestus bot already belongs to, one channel per call. Every scope it could not read is written down with the reason.
04Eval checkpoint
A judge reads the answer before you do. Any id, url, number, name or date that did not appear in a tool result during that run fails the answer, and the judge names the offending field so the search runs once more to repair that one thing.
05Condition
A high-confidence answer to a question somebody will ask again is proposed as one cited memory claim, quarantined until a person approves it. Everything else skips that branch and goes straight to the card.
06Output
The answer is filed, not just replied to: a document artifact with its own link and a comment on the ticket it belongs to. Nothing leaves Aestus — both agent steps are read-only and the workflow posts to no outside system.

By hand, and on Aestus

The same job, on either side.

By hand
You scroll a channel, then ask in it, then wait for whoever posted it to come back.
On Aestus
You ask once and get an answer with a link.
By hand
Keyword search finds the message, not the number inside the attachment.
On Aestus
The number is quoted with the cell, page or heading it came from.
By hand
Nobody notices the file was replaced two months ago.
On Aestus
A superseded file says so, names the file that replaced it, and names who last verified that.
By hand
You learn what the file is, never what it was for.
On Aestus
The card names the ticket, the plan and the decision the file fed.
By hand
"I could not find it" means nothing, because you do not know what was looked at.
On Aestus
Every answer lists the scopes, the time window and the queries that were run — most of all when nothing matched.
By hand
The next person asks the same question next month.
On Aestus
A high-confidence answer is proposed as a cited memory claim for a person to approve.
By hand
You have no idea what the search cost.
On Aestus
Every run is a row with its tokens, its latency and its cost.

What Aestus records

Every run is auditable after the fact, which is what makes handing the job over survivable.

The answer card
A document artifact with its own link and a revision history, marked with the run that generated it — and indexed, so the answer itself becomes findable.
What was searched
The scopes, the time window and the queries, kept on the card. A scope that could not be read is listed with the reason rather than left out, so a narrow search cannot read as a wide one.
The run
Tokens in and out, latency, cost, and a classified failure code when something breaks — on the same ledger as every other run in the workspace.
The proposed fact
A cited candidate memory claim, quarantined until a person approves it. An agent can never approve its own claim.

What it connects to

Only the tools this workflow actually uses. Each one has a setup guide.

slackOptional

Questions teams ask

Does it need a connection to work?

No. With nothing connected it searches your Aestus records — tickets, plans, projects, artifacts and approved memory claims — and the card says Slack was a scope it could not read. Slack widens the search; it does not enable it.

Can it read a channel I am not in?

No. It can only read channels the Aestus bot is a member of, and a channel belonging to a different Slack workspace returns nothing. The Slack connection is per person, so it reads as the person who asked.

What happens when it finds nothing?

It says nothing matched, lists every scope, query and time window it ran, gives up to three nearest misses and why each was rejected, and names one next step. That run is a success, not a failure.

Can it invent a number?

A judge fails the answer when any id, url, number, name or date did not appear in a tool result during that run, and names the offending field. The answer is rebuilt once before the run fails. Anything the reader cannot trace to a source is refused rather than softened.

Does it search my email?

No. Aestus has no read access to email, and this workflow reaches nothing outside Aestus and Slack.

All use cases How access is governed

Sources

lib/domain/search/searchAll.ts
lib/db/vocabulary.memory.ts
lib/domain/integrations/workflow-slack-policy.ts

Hand over Find anything your team has posted.

Tell us how your team does this today and we will set it up with you.

Talk to us

Questions teams ask

Does it need a connection to work?

Can it read a channel I am not in?

What happens when it finds nothing?

It says nothing matched, lists every scope, query and time window it ran, gives up to three nearest misses and why each was rejected, and names one next step. That run is a success, not a failure.

Can it invent a number?

Does it search my email?

No. Aestus has no read access to email, and this workflow reaches nothing outside Aestus and Slack.

Sources

lib/domain/search/searchAll.ts

lib/db/vocabulary.memory.ts

lib/domain/integrations/workflow-slack-policy.ts

The file someone shared in April, and the note that it was replaced

Find anything your team has posted

What you get

Found it — the enterprise tier number

How it runs

By hand, and on Aestus

What Aestus records

The answer card

What was searched

The run

The proposed fact

What it connects to

Questions teams ask

Other use cases

Hand over Find anything your team has posted.

The file someone shared in April, and the note that it was replaced

Find anything your team has posted

What you get

Found it — the enterprise tier number

How it runs

By hand, and on Aestus

What Aestus records

The answer card

What was searched

The run

The proposed fact

What it connects to

Questions teams ask

Other use cases

Hand over Find anything your team has posted.