Memory layer for your assistants
Your AI forgets. Construct doesn't.
Every decision, link, and insight you feed an assistant disappears when the session ends. Construct keeps it — classified, embedded, and searchable — then hands it back to Claude Code, Slack, your phone, or anything that speaks HTTP.
No card required · Capture from anywhere · Export everything, always
ConstructPersonal · 33⌘K search, ask, or captureai-tools15
last 5 Sep
productivity-tools12
last 15 Apr
automation8
last 4 May
communication-tools7
last 15 Apr
resource
The Field Guide to Building Products with AI
latecheckout.agency/ai-field-guide
ai-tools · automation
decision
Changed a channel to private, restricting access to invitation only
slack · #general
communication-tools
observation
Portal by Spotify cut my Claude Code token usage by 90%
x.com/rseroter
ai-tools · developer-tools
Developers in Claude Code
Decisions and gotchas from one session are there in the next, without pasting context back in.
Multi-assistant users
One memory layer behind every tool, instead of a different history in each.
Founders and operators
Why a call was made, months later, with the message it came from attached.
Ex-Notion, ex-Obsidian
No folders to maintain. Save it and let classification do the filing.
Capture in a second, recall in a sentence
Three stages run on every item that enters your brain, whether it came from a coding session, a Slack channel, or your phone's share sheet.
1Capture
Send a thought, a link, or a whole page from wherever you are. URLs get fetched and read for you.
web · mcp · slack · ios share · POST /api/entries
2Classify
Claude extracts a type, summary, people, topics, and suggested labels. Then it's chunked and embedded for search.
type · summary · labels · 1536-dim embeddings
3Recall
Ask a question and get an answer built from your own entries, with sources. Or let an agent query the brain directly.
semantic search · RAG answers · MCP tools
Six ways in, one place it lands
You shouldn't have to open an app to remember something. Construct takes input from wherever the thought actually happens.
MCP server
Claude Code and any MCP client can search, browse, and write to your brain mid-task.
claude mcp add construct --transport http --url .../api/mcp
Slack
Messages in a capture channel are ingested automatically. Query it without leaving Slack.
/brain search caching decisions
iPhone & Android
Share any link or selection from the share sheet. Two taps, saves first, classifies after.
Share → Construct → Save to brain
HTTP API
Anything that can make a request can write to your brain. Keys are scoped and revocable.
POST /api/entries { content | url }
Voice
Talk it out on the phone. The recording is transcribed, classified, and filed like anything else.
Hold to record → transcribe → save
Documents
Drop in a PDF and it's read, chunked, and embedded — searchable by the sentence, not the filename.
Upload PDF → chunk → embed
Answers with receipts
Ask in plain language. Construct searches semantically across everything you've saved and answers from those entries only — with each source listed, so you can check it or open the original.
Grounded only in what you saved — no invented sources
Every claim carries the entry it came from
Scope a question to one brain or search both
What did I decide about Slack channel access?
Construct · 2 sourcesOn 15 April you made a channel invitation-only, and logged it twice — once as the decision, once as the action taken.1,2 Both are labeled communication-tools.
1decision15 Apr
Changed a channel to private, restricting access to invitation only
2action15 Apr
Made a channel private so it can only be accessed by invitation
Labels you never have to write
Every entry is classified on arrival: a type, a summary, the people and topics in it, and suggested labels. Labels accumulate into clusters, so you can see what you actually think about instead of guessing at a folder name.
ai-tools 15productivity-tools 12automation 8communication-tools 7developer-tools 6learning-resources 5software-engineering 5developer-insights 4social-media 3trading-bots 3video-resources 3code-review 2hardware 2marketing 2
One brain, or one you share
Everything you capture lands in one brain, whatever account or tool it came from — filter by source when it matters. Shared spaces are separate by design, for a group that keeps re-deciding the same things.
Your brain
everything you saveevery sourceonly yours
Shared space
invite by emailshared clustersper-source permissions
You can belong to several shared spaces without any of it mixing into your own brain.
Pricing
Start free with your own API keys. Upgrade when you want Construct to handle the pipeline for you.
Free
$0/month
free forever
One brain, every capture source, everything searchable.
MCP server + API access
Slack and iOS capture
Auto-classification and labeling
Semantic search and Ask
ProMost useful
$—/month
price TBD
More than one brain, plus the tools that keep a library clean.
Everything in Free
Search and filter by source account or tool
Duplicate detection and label merging
PDF and document ingestion
Priority reprocessing of failed captures
Team
$—/user/month
price TBD
A shared brain for a group that keeps re-deciding the same things.
Everything in Pro
Shared clusters and entries
Per-source permissions
Shared Slack capture channels
Audit trail on decisions
SSO
Prices aren't set yet — the placeholders above are where they'll go. Free stays free.
Questions
Where does my data live?
In Postgres with pgvector — your entries, summaries, labels, and embeddings. You can export all of it at any time.
What happens if a capture fails?
The entry is still saved with its URL and content. It's marked unprocessed so you can retry it later — a failed fetch never loses the link.
Can I use it with tools other than Claude?
Yes. The MCP server works with any MCP-capable client, and the HTTP API works with anything else.
Can I export my data?
Always, on every plan. Entries, labels, and summaries export as JSON or Markdown.
How does search actually work?
Entries are chunked and embedded, then retrieved by vector similarity. Answers are generated from the retrieved chunks only, with the source entries listed.
Can I keep work and personal separate?
Your own captures all land in one brain, tagged by the source they came from, so you can filter rather than switch. Anything genuinely shared lives in a shared space, which never merges into your personal brain.
Do you use analytics?
Yes, Google Analytics, which sets cookies and records which pages are viewed. Page addresses are stripped of ids and search text, and no memory or chat content is ever sent. See Google's privacy policy (https://policies.google.com/privacy) or opt out with the browser add-on (https://tools.google.com/dlpage/gaoptout).
Start remembering what you already knew
Connect an account, paste one MCP command, and the next thing you tell Claude stays yours.