Science comes first.
Companies come last.

Tibeb is an AI research terminal for new technology. A thesis starts with what became possible — a paper, a trial, a standard — and says in advance what the filings would show if it is right. It goes through the diligence an investment committee would ask for, and only a thesis that holds up gets to the companies exposed to it.

Sign in See the eight checks

Access is by approval. Apply for access — approved readers start on Free, with $2 of research a month. Firms: the API and a pilot.

  1. Discover

    Two lanes across thirty-two themes, from AI memory and lasers to nuclear and neo banks: what became possible — new papers and what builders are arguing about — and what was filed or said, with the narratives moving the market: who made the claim, how far the shares moved, who answered.

    • AI hyperscalers
    • AI memory
    • Photonics
    • Lasers
    • Neo cloud
    • Cybersecurity
    • AV & robotics
    • Nuclear
    • Space tech
    • Crypto
    • Healthcare tech
    • Macro

  2. Thesis

    One yes-or-no question with a date, and what you would expect to see either way — written down before the evidence is gathered, so the filings can prove it wrong.

    An example

    Will a merchant switch ship with co-packaged optics in volume by the end of 2027?

    Resolves by 2027-12-31

  3. Diligence

    Eight checks on the thesis and on the sources behind it: the result it starts from and the path to revenue, the filings and papers it cites, the case against it, a base rate, days it can be wrong on, six seats that argue from the ledger.

    • Market expectation
    • Mechanism mapped
    • Falsifiable thesis
    • Primary evidence, both ways
    • Sources verified
    • Base rate attached
    • Settling catalyst
    • Committee verdict

  4. Exposure

    The companies the evidence shows are exposed, how, and how much of each business turns on the thesis — with the share of revenue read from the company's own 10-K.

    • Direct exposureBenefits
    • Supplier or bottleneckBenefits
    • Second-orderBenefits
    • At riskHurt

What a thesis has to get through

Before the list of exposed companies opens, the thesis and its evidence pass eight checks — the questions an investment committee asks. They are the engine's, computed on the server, the same for everyone. Tibeb works your thesis through them after deep research, and asks you wherever a person should decide.

  1. Market expectation

    What the market already expects about the question — an analysts' consensus, the company's own guidance, an industry's published schedule — quoted from a source on file, and whether a yes agrees with it, goes further, or runs against it.

    A thesis the market already expects can be right and still tell you nothing new. Knowing which comes first.

  2. Mechanism mapped

    How the claim gets to revenue, step by step — from a measured result to a device, a product a customer qualifies, volume and sales — with how long each step usually takes, and which one holds everything up. A question that starts in the market traces a money map instead: who pays whom, in four links, from a share price measured against its benchmark.

    A result that cannot reach revenue is science, not a thesis.

  3. Falsifiable thesis

    The thesis is a yes-or-no question with a full date. It names the document that will settle it, and what that document would show if the answer is yes and if it is no.

    If nothing could prove it wrong, there is nothing to research.

  4. Primary evidence, both ways

    At least 3 sources that are the thing itself — a paper, a trial, a patent, a standard, a dataset, a company's filing or a regulator's own document; a question one company's filings will settle needs three filings. And both sides: at least 1 source of its own arguing for the thesis and 1 against it, the strongest counter-argument there is. A reprinted press release is not its own publisher, and a rumor counts for nothing.

    Three documents that decide the question beat twenty articles about it. The case against is gathered on purpose.

  5. Sources verified

    Every source is opened and read, so a source with no link can't count, and a figure is kept only when its page states it. When fewer than 80% of the links still open, the check says so: a dead page is flagged, not counted against the thesis.

    A link only shows that a page exists. The quotation shows that the page says it.

  6. Base rate attached

    How often similar questions have come out yes, counted from past cases on a page that states the figure — how many cases, and how many were yes. The model does not work the rate out itself.

    Before the evidence is weighed, ask how often this kind of thing happens.

  7. Settling catalyst

    At least 1 dated event on an official page that settles the question, with who owns it. What you would expect to see if the thesis is true and if it is false is written down before the evidence is weighed. Other dated events on the way are kept, to re-check the thesis as they pass.

    The thesis names the day it could be proved wrong, and what it expects that day to show, before anyone weighs the evidence.

  8. Committee verdict

    Six readers each give the chance the thesis comes out yes, from the evidence in front of them, and none of them is shown an earlier answer. The server turns those chances into one figure; the model does not write the number. One reader argues against the thesis and is reported on their own, not folded into that figure. If the readers are far apart, there is no verdict.

    The list of exposed companies opens after this verdict, and each company on it comes from its own filings.

Change the question and the review, the verdict and the exposure go stale. They have to be earned again for the thesis you are actually holding.

Science first, filings as the test

A Tibeb thesis starts with a measured change in what is physically, biologically or computationally possible. Filings and news confirm or deny what it predicted. A company is named only after the gate.

  1. Start with what became possible.

    A paper, a trial readout, a patent, a standard, a regulator's review — the advance stated in units, not adjectives. A filing, a catalyst or a move in the shares can open a question too, and the thesis says where it came from.

  2. Predict before you look.

    The thesis says what you would expect to see if it is true and if it is false, before the evidence is gathered. The filings then confirm or deny it: the pre-registration a physician recognizes, the falsifiability an engineer expects.

  3. Companies come last, from their own filings.

    Exposure names the companies the evidence shows are exposed: the role, the direction, how much of the business turns on the thesis, and the share of revenue in the 10-K's own words. Nothing is estimated to fill a gap.

Also in the terminal. Each thesis is marked by when its question settles: within a year, between one and three years, or later, so a question due next quarter is not judged the same way as one due in two years. When the obvious company's share price already reflects the thesis, Tibeb looks one step along the suppliers and customers that company's filings name, and stops there. Each of the thirty-two topics has a short brief of the field's limits and benchmarks, and it stays a draft until a person has checked it and signed it. As your theses reach their dates, each verdict is kept next to what actually happened, including the ones that were wrong.

Beyond the four steps

The same account, in a few more places.

  1. Today and the recap.

    A home page of what changed in your theses since you last looked and what waits on you, and a recap that watches the companies your theses name for sharp moves, checking each move's story against the filings.

  2. The desk.

    A page for each company in the universe, with its SEC filings, TradingView's live charts and the market map, beside the theses that name it.

  3. The apps.

    Apps for iPhone, Android and Apple Vision Pro read the same theses and the same feed. They aren't in the app stores yet; the web terminal has everything they do.

It stops at research

The terminal ends at the companies exposed to the thesis: how it reaches each one, in which direction, and how much of the business turns on it, with the sources behind every line. What you do with that is your call.

Nothing in Tibeb is investment advice.

  • orders placed
  • position sizing
  • buy or sell calls
  • price targets

Pricing

Every reader is approved by hand and starts on Free. Each plan includes a month of research, measured at what the AI models cost us to run it (prorated by days the month a plan starts or changes); past it, a paid plan's prepaid credit keeps you going. Plans renew until you cancel, and a cancellation takes effect at the end of the period you paid for.

Free

$0

  • $2 of research a month
  • Discover, theses and diligence
  • No card needed
Apply for access

Pro+

$89/month

  • $40 of research a month
  • 3 API keys and the MCP server
  • Agent runs through the API, with approvals and signed webhooks

Founding member: Pro+ for a year at $499, 150 places, while open.

Apply, then choose Pro+

Team

$300 a seat/month

  • 3 seats at least, one bill
  • Pro+ for every seat, one shared allowance of $40 a seat
  • 10 API keys a team; admins invite members, who take a seat by accepting
  • Shared team workspaces for theses: planned, not built yet
Request a team Email hello@tibeb.trade

Past a paid plan's allowance, research draws on prepaid credit at 2× the AI models' cost: packs of $10, $25 or $50, and, if you turn it on, an automatic top-up when credit falls below $3 (off, Tibeb stops at $0). Free doesn't buy packs. A bigger plan or more seats are charged at once for the rest of the period; a smaller one starts at the next. Cards are handled by Stripe and never reach Tibeb's server. What a firm gets · Security

For firms

Everything a reader sees on a thesis — the question, both scenarios, the eight checks, every source with the side it argues and the page's own words, and the exposed companies once the gate opens — can be read through a REST API, or by your AI tools through an MCP server.

  • Keys you make under Tools → API keys, shown once and stored only as a hash. Each key reads one login's records.
  • 120 calls a minute a key, and every call written to the key's audit log: when, the route, the answer's status.
  • Agent runs work a thesis through the engine's stages. The lock waits for a person's approval, and each step keeps what it did and what it cost.
  • Signed webhooks when a run completes, fails or waits: HMAC-SHA256 over the body, sent only to https addresses on the public internet.
  • Research only. No orders, no buy or sell calls, and nothing starts a model call except an agent run you start.
Read the API docs Security
List your theses, then ask the MCP server what it offers
# REST
curl https://www.tibeb.trade/v1/theses \
  -H "Authorization: Bearer tbk_…"

# MCP (JSON-RPC 2.0 over HTTP)
curl https://www.tibeb.trade/mcp \
  -H "Authorization: Bearer tbk_…" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

# → list_theses, get_thesis, ticker_filings,
#   start_run, get_run

Apply for access

Every reader is approved by hand. Once you are, you'll get an email to set your password, and you start on Free: $2 of research a month, no card needed.

  • Want more? Choose Pro ($29/month), Pro+ ($89/month) or a Team plan on the billing page once you're in.
  • Past the allowance, a paid plan draws on prepaid credit at 2× the cost of the AI research it runs: packs of $10, $25 or $50, and an automatic top-up if you turn it on. Free doesn't buy packs: its allowance starts again each month.
  • Cancel any time. A plan renews until you cancel; it holds to the end of the period paid for. Pricing