Why we're building Astell, and what we believe a company actually is.
Every company has a memory. Right now, in yours, it's a person. A founder, a chief of staff, a PM.
The decisions, the promises, the "we said we'd do this" live in their head, their inbox, a Slack DM nobody else can search.
By fifty people, the person can't hold it all.
MAR 14
Notion·Maya call notes
Maya
“Can you have the custom export ready by July? Our finance team needs it for the audit.”
You
“Yeah, July should work.”
Linear·MAYA-204 created
“Custom export for Maya's team”
Backlog · No priority
MAY 6
Google Calendar·Engineering planning
Q3 roadmap reviewed.
MAYA-204 stays in backlog.
eight weeks
JUL 1
GitHub·PR #487 merged
“Refactor export pipeline”
Gmail·Maya
Maya
“Quick check on the export next week. Our auditors arrive Monday.”
Slack·You → Priya
You
“Did we ever commit to this?”
Slack·Priya
Priya
“Ticket's been in backlog since March. Nobody flagged it.”
eleven weeks
SEP 16
Gmail·Maya
Maya
“We're going to take a step back for now. Thanks for the partnership.”
Hindsight is 20/20
A loop is a decision, promise, or request that hasn't reached where it needs to live.
The pieces are usually all there. They just don't meet in time. Close the loops still open. Surface the ones about to break.
Astell draws the arrows while the loop is still open.
Loops dashboard
The view you wished you had in May.
This is Maya's loop, rendered in Astell. Same artifacts. Same timestamps. The arrows already drawn. Status flagged in May. Months before the goodbye email.
Loop activity
+3net · 30d
Loops
9
Live
At riskBreaks in 8wPromise
Custom export for Maya's team
Synced 2m ago
Artifacts4
Tools4
Days open53
DeadlineJul 1
Maya asked for the export by July for her audit. MAYA-204 sits in backlog with no owner. The Mar 14 Slack thread went quiet after a day. No motion in 53 days. Auditors arrive in 8 weeks.
People in this loop
Suggested next steps
Citations
Sourced to the sentence that proves it.
export-pipeline-spec.pdf
2/ 3
Page 2 of 3
3.2 Export deliverables
The data platform team owns CSV and Parquet export pipelines. All customer-facing exports must pass schema validation before release.
Audit customers require a dedicated export format with fixed column ordering and signed checksum metadata attached to each file.
Custom CSV export must be ready before audit week.
Finance will validate sample output two weeks prior. Engineering must assign an owner and confirm the delivery date in Linear.
See appendix B for column definitions and Finance sign-off checklist.
Unified search
Everything your team knows. Every tool you use.
Maya export
Top results · 8
Work anywhere
One ask. Astell gathers the context and does the work.
Or run from
Update MAYA-204 to close the Maya export loop. Assign Priya and set a real due date before it breaks.
Searched Notion, Linear, Gmail & Slack
This loop breaks in 8 weeks. Confirm the update and I'll push it to Linear:
Update issuevia Linear
Confirmation required
Custom export for Maya's teamPromiseAt riskCustomer
Loop at risk. Auditors arrive in 8 weeks. Mar 14 call: Maya asked for a custom export by July; ticket sat in backlog since March. Priya flagged it in Slack this morning.
To close the loop: owner → Priya, due → Jul 1, link Notion call notes.
Retrieval is what every "AI assistant" does. It fetches your data every time you ask. Live calls to Slack, Gmail, Notion. Slow. Hit or miss. No memory of what came before.
Ingestion is the opposite. We pull your data in once and keep it in sync. Every artifact lives in three places at the same time.
How Astell turns your tools and artifacts into living, permission-aware memory that your team and any agent can use.
Your toolsTools & every artifact type
Continuous ingestAlways on
Artifact-aware pipeline
OCR & transcribe
Parse & chunk
Normalize
Resolve & dedupe
Enrich context
Permissions
Unified memory
RelationalStructure
VectorMeaning
GraphLinks & time
Permission-awareFiltered by identity
Put to work
Your team
Loops
Notifications
Any agent
Comms
Tools & web
1
In a relational store, so we know the ticket is a ticket and the email is an email. That's where structure lives.
2
In a vector store, so we know "Maya's check-in" and "the audit follow-up" are about the same thing. That's where meaning lives.
3
In a graph, so we know which thread led to which ticket led to which decision. That's where the shape of a loop lives.
Search tools live in vectors. Databases live in relations. Knowledge graphs live in graphs. Nobody else holds all three at once, so nobody else carries the full shape of a loop.
The result is queries that return in under a second, with the shape already there.
We built the ingestion engine for ourselves. It turned out good enough that other teams now pay us to run theirs on it too.
And time. The Linear ticket renamed in March. The owner who handed it off in May. The deadline that moved twice. Most systems overwrite. Astell remembers.
Permissions are not bolted on. We mirror them from your source tools in real time, identity by identity. What you can see in Slack, in Gmail, in Notion, you see here. Nothing crosses lines.
The same context is available to your team, to Astell, and to whatever agent you bring. Claude. ChatGPT. Your own. The loop doesn't get rebuilt every time you switch tools.
We see a future where Astell is part of every team's workflow, just like Slack, Notion, and Google.
