Om-E · intelligence dashboard
Everything you've ever read, ready to answer.
Om-E is an Agentic AI living inside your Google browser. Om-E takes the chaos of life's unstructured data — receipts, PDFs, transcripts, .md's, plus multiple formats including images — and dynamically converts everything into structured vector stores that live locally on your physical disk. All files are converted to markup, stored and administered by you and your new best friend…
macOS · Chrome extension · local-first
How this works
Open tab to cited answer, in six moves.
scroll to walk the loop ↓
World first
The world's first AI context engineering tool.
Every answer is only as good as the context it was given. Om-E makes that window yours to run — attach a KB for retrieval, load a topic in full, or send the question bare. What the model sees, what it ignores, and what you pay: dialled per turn.
Attach vs load
Retrieval-only when the KB should answer, full text when the model must read every word.
Scoped per chat
Each chat sees the knowledge you point it at — nothing else bleeds in.
Cheaper by design
Tokens are the bill. Sending less, better-chosen context is the discount.
Sharper answers
Less noise in, less noise out — the model reasons over what matters, not everything.
The machine
Four moving parts. One workspace.
01Chats
Where the work happens.
Every conversation saved, searchable, resumable — nothing you build in a chat is ever lost.
01 Attach or load — two gears
Attach a topic and local RAG surfaces the relevant passages. Load it and the full text is in context.
02 Where topics meet
Join several sources in one thread and weigh them against each other — vendor vs vendor, draft vs brief.
03 Cited and verified
Grounded answers cite the topics they drew on, and verification reports are built right in the thread.
04 Messages you can shape
Edit or delete any message — or have the AI clean a thread: extract the good parts, format them, distil raw turns into a bio.
02Topics
The unit of memory.
Anything worth keeping, saved once and reused forever — drop a topic into any chat, any time.
01 Born from anything
File imports, captured pages, transcripts, chat answers, plain notes — one shape for all of it.
02 Images are searchable too
Import a document and the images inside are vectored alongside the text — the diagram surfaces in the same search as the words.
03 Provenance built in
Where it came from and when it was captured travel with the topic for good.
04 Even formats are topics
A topic can define how you want output to look — attach it and answers arrive in your structure, your style.
03Knowledge Bases
Libraries with meaning.
Libraries of topics that search by meaning — your private index, on your machine.
01 One memory per job
A KB per client, project or domain — deliberate, curated containers, not one big pile.
02 Search by meaning
Vector retrieval finds related ideas and passages even when the wording is different.
03 Scoped answers
A chat attached to a KB draws on that KB alone — client work never leaks into policy work.
04 Search one, or search all
Scoped to a KB by default — or sweep every KB at once when you can't remember where you put it.
04Multi-session
Everything runs at once.
Fire a research run in one chat and keep working in another — the machine never makes you wait.
01 Truly concurrent
Chats run side by side in their own session spaces — a long run in one never blocks another.
02 Isolated by design
Each session keeps its own context, persona and knowledge scope — nothing bleeds across chats.
03 Fire and move on
Kick off a research run, switch chats and keep working — the run carries on and lands when it's done.
04 Every thread keeps its trail
Switch back any time; each session resumes exactly where it was, sources and reasoning intact.
What makes Om-E different
Answers from your world.
Most AI tools answer from a blank box. Om-E is built around context engineering: you stay in control of the sources, documents, pages, transcripts and notes the AI works from — so knowledge compounds instead of vanishing after one chat.
Context-controlled
You decide what the model sees on every turn — attach knowledge for retrieval, load it in full, or keep the chat clean.
Knowledge-preserving
Work saves as Topics inside local Knowledge Bases — searchable by meaning, reusable across every future job.
Evidence-based
Citations, verification rubrics and A|B comparison — judge answers by substance, not vibes.
Local by design
The index lives on your machine. Your knowledge isn't trapped inside a platform you can't control.
Built for trust
Answers you can hand to someone else.
Everything Om-E hands you carries its receipts — where each claim came from, how it scored against your rubrics, and how the sources stack up. Ship the work; the evidence travels with it.
Citations
Sources travel with the answer — whoever you hand it to can check it themselves.
moreVerification
Score sources against your own rubrics — evidence, not vibes.
moreComparison
Source vs source on your criteria, laid out where each one wins.
moreTemplates
Turn captured research into finished documents in your format.
more
Built for trust
Judged by substance, not vibes.
01Citations
Receipts included.
Sources travel with the answer — whoever you hand it to can check it themselves.
01 Claims point home
Every claim in the answer carries a link back to the topic it came from.
02 Provenance, not vibes
The receipt names the document, the section and the day it was captured.
03 Check it yourself
Open the cited topic and read the original source — no trust required.
04 It survives the handover
The person you send it to can verify every line without you in the room.
02Verification
Judged, criterion by criterion.
Score answers against rubrics you define — the judgement is inspectable, not vibes.
01 Your criteria
The rubric is yours — what counts as good is defined by you, not a black box.
02 Criterion by criterion
See exactly where a source passes and where it falls short — no single mushy score.
03 Inspectable judgement
Every score cites the passage that earned it, so the call can be challenged.
04 Reusable rigour
The rubric saves as a topic — run the same bar over the next vendor too.
03Comparison
A or B, decided.
Two proposals, two policies, two vendors — where they differ, who wins on which criterion.
01 Real differences only
The noise both documents share is skipped; what actually diverges surfaces.
02 On your criteria
Price, security posture, retention terms — the comparison runs on axes you pick.
03 Claims cite their side
Every difference points to the exact spot in each source that proves it.
04 It saves as a matrix
The comparison lands back in your KB as a topic — reusable in the decision memo.
04Templates
Never start blank.
Turn captured research into reports, briefs and memos in your format — citations intact.
01 Your format
Templates you define — house style, required sections, the order your org reads in.
02 Fed by your KB
The substance comes from captured knowledge, not the model's general guesses.
03 Citations survive
The finished document keeps its receipts — claims still point to their sources.
04 It round-trips
Outputs save back as topics: this quarter's memo is next quarter's source.
Multi-session
Nothing waits for anything.
Each chat is its own session: its own context, its own knowledge scope, its own running work. Fire a deep research run in one chat, draft in a second, triage in a third — they run side by side, and every thread keeps its own trail.
Truly concurrent
Chats run side by side in their own session spaces — a long run in one never blocks another.
Isolated by design
Each session keeps its own context, persona and knowledge scope — nothing bleeds across chats.
Fire and move on
Kick off a research run, switch chats and keep working — the run carries on and lands when it's done.
Every thread keeps its trail
Switch back any time; each session resumes exactly where it was, sources and reasoning intact.
Work with it
Point it at the job. It works.
01Agentic
It does, not just says.
It browses, captures, drafts — and saves the work back into your KB.
01 Hands, not just answers
It drives the browser, runs the searches and does the capturing — not just the talking.
02 The work round-trips
Everything it drafts or captures saves back into your KB as topics — no copy-paste step.
03 Tools that grow
New capabilities plug in through MCP — the agent extends without waiting for a new app.
04 You hold the reins
Watch the run, stop it mid-errand, redirect it — the agent works for you, visibly.
02Context engineering
The dial is yours.
A context dial on every turn — what the model reads, and what you pay for.
01 Attach vs load
Retrieval-only when the KB should answer, full text when the model must read every word.
02 Scoped per chat
Each chat sees the knowledge you point it at — nothing else bleeds in.
03 Cheaper by design
Tokens are the bill. Sending less, better-chosen context is the discount.
04 Sharper answers
Less noise in, less noise out — the model reasons over what matters, not everything.
03Personas
Pick the eyes.
Review it as a researcher, a governance officer, a lawyer — pick the voice per chat.
01 Picked per chat
Each chat runs under the brief that fits its job — switch chats, switch eyes.
02 Style with substance
Not a tone slider — what the agent prioritises, checks and flags actually changes.
03 Same KB underneath
No re-importing, no copies — every persona reads the same captured knowledge.
04 The review multiplier
Run the same question under three personas and you've had three reviewers, not one.
Deep research
Multi-source research runs that come back as sourced, finished reports.
moreWeb search
Built-in search — the agent searches, opens the tabs and reads them for you.
moreAgentic
An agent with tools, not a textbox — it browses, captures, drafts, and saves the work back into your KB.
morePersonas
Same knowledge, different eyes — review it as a researcher, a governance officer, a lawyer.
more
Capabilities
The agent goes out and gets it.
01Deep research
Ask big. Get a report.
Multi-source research runs that come back as sourced, finished reports — not a pile of links.
01 Hours become one prompt
The searching, opening, reading and collating happen without you holding the tabs.
02 Finished, not fragments
What comes back is a written report — structured, readable, ready to hand over.
03 Receipts attached
Every claim points back to the source it came from, with capture provenance.
04 It lands in your KB
The report saves as a topic — retrievable, comparable, reusable in the next job.
02Web search
It goes and finds out.
Built-in search — the agent searches, opens the pages and reads them for you. Sources attached.
01 No key setup
Search is built into the agent — there's no third-party search key to configure.
02 It reads, you don't
The agent opens the result tabs and does the reading; you get the findings.
03 Sources attached
What comes back is cited — you can check any finding against the page it came from.
04 Keep what's useful
Capture the good pages as topics and they join your KB — still there when the site changes.
Let it loose
When your KB can't answer, it goes and finds out.
Built-in web search and deep research runs — the agent searches, opens the results, reads what's worth reading, and what comes back lands in your KB as knowledge.
Local-first
Nothing uploads. Nothing expires.
Your KBs, files and keys live on your machine — the index is local, retrieval is local, and the only thing that ever leaves is the turn you chose to send.
how we handle your data →Tokens, not seats
Buy a pack. Spend it on turns.
No subscription, no per-seat maths. Credits ride a hard-capped key, sharper context costs less per answer, and the meter is always yours to read.
see the packs →Stop losing knowledge in tabs.
Ask the page. Save the answer. Build your knowledge base.