Tibeb's API

GET/v1/theses

The key's login's theses (up to 200): the question, the resolve-by date, the status, how many of the eight checks pass, and the committee's number when there is one.

{
  "theses": [
    {
      "id": "w1",
      "name": "AAOI",
      "question": "Will AAOI report $300M revenue for a quarter by 2027-03-31?",
      "resolve_by": "2027-05-31",
      "status": "open",
      "updated": "2026-11-08T14:02:11Z",
      "gate": {"passed": 3, "total": 8, "open": false, "whose": "engine"},
      "verdict": null
    }
  ],
  "version": "2026-11-04"
}

gate.whose is engine once the thesis is in the research engine, else web (the provisional checks). verdict, when present, is {"p_yes": 62.0, "lean": "yes"} — the committee's pooled probability, a research estimate, not a forecast of any price.

GET/v1/theses/{id}

One thesis in full: both scenarios, each of the eight checks, every source with the side it argues and the page's own words, the dated events — and the exposed companies only once all eight checks pass.

{
  "id": "w1",
  "question": "Will AAOI report $300M revenue for a quarter by 2027-03-31?",
  "resolve_by": "2027-05-31",
  "resolve_by_text": "a 10-Q, by 2027-05-31",
  "status": "open",
  "gate": {"passed": 3, "total": 8, "open": false, "whose": "engine"},
  "scenarios": {"yes": "InP supply loosens.", "no": "Supply stays tight."},
  "checks": [
    {"id": "1", "label": "Market expectation", "pass": false, "detail": "…"}
  ],
  "evidence": [
    {"id": "s1", "title": "AAOI 10-Q", "url": "https://www.sec.gov/…", "publisher": "",
     "date": "", "category": "filing", "side": "for", "link_checked": true,
     "passage": "Revenue was $192M."}
  ],
  "catalysts": [{"date": "2026-11-05", "title": "Q3 report", "url": ""}],
  "exposure": [],
  "exposure_note": "Companies are named once all eight checks pass."
}

side is for, against or neutral. Once the gate is open, each exposure row is {symbol, company, role, direction, scenario, exposure, reason}.

GET/v1/tickers/{symbol}/filings

The company's ten latest filings, from the SEC's own list for it. A symbol that isn't an SEC filer answers "filer": false with no filings.

{
  "symbol": "AAPL",
  "filer": true,
  "cik": 320193,
  "filings": [
    {"date": "2026-10-31", "form": "10-K", "company": "Apple Inc.", "cik": 320193,
     "url": "https://www.sec.gov/Archives/edgar/data/320193/…/aapl-20260927.htm",
     "path": "edgar/data/320193/0000320193-26-000105.txt", "description": "10-K"}
  ],
  "source": "SEC EDGAR"
}

GET/v1/pe/lbo/prefill

?symbol=MU or ?cik=723125. A US SEC filer's latest fiscal year from its XBRL — revenue, operating income, D&A, capex, debt, cash and shares outstanding — each with its tag, period, form, filing date, accession number and the filing's link. EBITDA is operating income + D&A, and says so. A figure the filing doesn't give is listed in missing, never estimated. Market cap and price are licensed market data and are not served until Tibeb's display licence is in place (market.licensed: false).

POST/v1/pe/lbo

The model: sources and uses, the yearly projection, the debt schedule with mandatory amortization and the cash sweep (senior first), leverage and interest coverage each year, exit equity, IRR and MOIC, and two sensitivity grids (entry × exit multiple, senior leverage × exit multiple). No model call. A target is a filer ({"kind": "filer", "symbol": "MU"}, its figures from the SEC; a figure in figures stands over the filing's and is marked as yours) or a private company ({"kind": "private", "name": …} with figures.revenue and figures.ebitda). Percentages are in percent. Inputs that can't make a model answer 422 with each problem in words.

curl https://www.tibeb.trade/v1/pe/lbo \
  -H "Authorization: Bearer $TIBEB_KEY" \
  -H "Content-Type: application/json" \
  -d '{"target": {"kind": "private", "name": "Acme Packaging"},
       "figures": {"revenue": 500000000, "ebitda": 100000000},
       "assumptions": {"entry": {"mode": "multiple", "multiple": 9},
         "senior": {"multiple": 3.5, "rate": 8, "amort": 5}, "sub": {"multiple": 1.5, "rate": 11, "amort": 0},
         "sweep_pct": 75, "transaction_fee_pct": 2, "financing_fee_pct": 2.5, "min_cash": 20000000,
         "years": 5, "growth_pct": [6, 6, 5, 5, 4], "margin_pct": 20,
         "da_pct": 3, "capex_pct": 3.5, "nwc_pct": 10, "tax_pct": 25, "exit_multiple": 9}}'

{
  "sources_uses": {"uses": {"enterprise_value": 900000000.0, "total": …}, "sources": {"sponsor_equity": …}},
  "years": [{"year": 1, "revenue": …, "ebitda": …, "fcf": …, "senior_end": …, "leverage": …, "interest_coverage": …}, …],
  "returns": {"equity_invested": …, "equity_at_exit": …, "moic": …, "irr": …, "years": 5},
  "sensitivity": {"entry_exit": {"rows": [7, 8, 9, 10, 11], "cols": [7, 8, 9, 10, 11], "irr": [[…]], "moic": [[…]]}, …},
  "provenance": {"revenue": {"kind": "user", …}, …},
  "memo_ceiling_usd": …
}

POST/v1/pe/lbo.xlsx

The same body; the model as an Excel workbook — sheets Inputs, Model, Debt, Returns, Sensitivity and Sources. Every cell on Model, Debt and Returns is a live formula over Inputs, so changing an input recalculates the model. Each figure from the SEC carries a comment with its tag, period and filing, and a row with the link on Sources; your assumptions are marked as yours. The sensitivity grids are values (a grid of whole models needs a data table).

POST/v1/pe/ic-memo

{"lbo": <the body above>, "thesis_id": "w1", "with_10k": true}. One model call, booked to your key's login. The memo's sections — situation, business, thesis and value creation, financing and returns, key risks and mitigants, diligence questions, what would change the view — from the LBO's numbers, the SEC's figures, your thesis's evidence and passages of the latest 10-K. Code then checks it: a sentence with a figure that isn't in that material, a quote that isn't in the filing or a source's own words, or any recommendation wording is removed and listed under removed. ?format=docx or ?format=md returns the file; the JSON carries markdown and docx_base64.

GET/v1/pe/prefill

?symbol=MU or ?cik=723125 (&what=statements, the default; what=lbo is the LBO's shorter read). The income statement, balance sheet and cash-flow statement for the latest three fiscal years (annual) and the latest quarters the filings report (quarterly), the cover's shares outstanding, a base case the filings' averages suggest for the three-statement model (suggested_drivers, each with how it was averaged — an offer, used only when you send it back), the 10-year Treasury par yield on file for the DCF's risk-free rate (risk_free_on_file), and the market cap and price for a licensed caller.

POST/v1/pe/three-statement

The base year from the filing (target: {"kind": "filer", "symbol": "MU"}; a line in figures stands over it) or typed ({"kind": "private", "name": …} with the base-year lines), years (3–10, five by default) and scenarios — a set of drivers per case (base, bull, bear, any names): growth_pct (one, or one a year), gross_margin_pct, sga_pct, rnd_pct, da_pct (da_basis revenue or ppe), capex_pct, dso, dio, dpo, tax_pct, interest_pct, revolver_pct, payout_pct, min_cash, and optionally debt_change and buybacks by year. Interest is on opening balances (nothing circular), the revolver is the plug and assets equal liabilities plus equity every year. Each case's statements, ratios and a side-by-side summary.

curl https://www.tibeb.trade/v1/pe/three-statement \
  -H "Authorization: Bearer $TIBEB_KEY" \
  -H "Content-Type: application/json" \
  -d '{"target": {"kind": "filer", "symbol": "MU"}, "years": 5,
       "scenarios": {"base": {"growth_pct": 8, "gross_margin_pct": 40, "sga_pct": 12, "rnd_pct": 10,
         "da_pct": 9, "capex_pct": 15, "dso": 45, "dio": 120, "dpo": 50, "tax_pct": 15,
         "interest_pct": 5, "revolver_pct": 7, "payout_pct": 10, "min_cash": 2000000000}}}'

POST/v1/pe/dcf

flows: {"source": "three_statement", "three_statement": <the body above>, "case": "base"} or typed lists (ebit, da, capex, nwc_change, optionally ebitda, and tax_pct); wacc: risk_free_pct (left out, the Treasury yield on file is used and said so), beta, erp_pct, cost_of_debt_pct, tax_pct, debt_weight_pct; terminal: growth_pct and optionally exit_multiple; mid_year; bridge (debt, cash, minority interest, preferred stock, diluted shares — the filing's unless typed); price (typed; a licensed caller's is Massive's). Unlevered free cash flow by year, both terminal values, EV and equity, the implied price, sensitivity grids over WACC × growth and WACC × exit multiple, and — with a price — the reverse DCF: the terminal growth, and the uniform growth of free cash flow, the price implies.

POST/v1/pe/comps

peers (tickers, up to 15), an optional target, market_caps by symbol (typed; a licensed caller's come from Massive) and figures that stand over the filings'. Each company's latest fiscal year — revenue, EBITDA (operating income + D&A), net income, debt, cash, diluted shares — EV / revenue, EV / EBITDA and P / E, the count, mean, median and quartiles (QUARTILE.INC), outliers by 1.5 × IQR, the target's implied range, and suggested: filers on the target's SIC line, suggested, never applied on their own.

POST/v1/pe/deal-comps

deals: each with target, acquirer, date, ev, revenue, ebitda, premium_pct and your source note (no free feed carries deal terms); a deal with no ev may name its 8-K (filing: {"cik", "accession"}) and the value is read from the filing's own words, kept only when a sentence states the figure, and quoted. figures for the target (or a filer's from the SEC). The multiples, the statistics, the premiums and the implied range.

POST/v1/pe/merger

acquirer and target (filers from the SEC, or private with figures), deal: offer_price or premium_pct (over target_price), stock_pct, new_debt_pct, acquirer_price, new_debt_rate_pct, cash_yield_pct, synergies (pre-tax, by year), fees, tax_pct, years (1–5) and optional net income growth for each side. Pro forma EPS and accretion / dilution by year, breakeven synergies, a stock-mix sensitivity, contribution analysis and a pro forma balance sheet with goodwill (purchase price − book equity; no write-ups, stated).

POST/v1/pe/three-statement.xlsx

POST/v1/pe/dcf.xlsx

POST/v1/pe/comps.xlsx

POST/v1/pe/deal-comps.xlsx

POST/v1/pe/merger.xlsx

The same bodies; each model as an Excel workbook of live formulas (Inputs and the model's sheets, the sensitivity grids as the engine's values, and Sources). A figure from the SEC is commented with its tag, period and filing; a figure a licence withholds is an empty cell with a comment saying why.

GET/v1/rates/curve

?date=2026-11-06 (optional): the newest day on file on or before it. The par points (each tenor in years), the same a week, a month and a year before, the zero curve (compounded semiannually) with each node's discount factor, the one-year forward from every half year, the named forwards (1y1y, 2y1y, 3y2y, 5y5y, 10y10y, 20y10y), the measures — 3m10y, 2s10s, 5s30s and the 2s5s10s butterfly in basis points from the Treasury's own points, the 5- and 10-year breakevens in percent (nominal less real par) — the real curve and the bills. Bills out to a year are read on the Treasury's bond-equivalent basis; from a year out the par bonds are bootstrapped on a half-year grid, between the Treasury's points by a monotone cubic (PCHIP).

curl "https://www.tibeb.trade/v1/rates/curve?date=2026-11-06" \
  -H "Authorization: Bearer $TIBEB_KEY"

{
  "day": "2026-11-06",
  "par": [{"label": "3M", "t": 0.25, "rate": 3.88}, …, {"label": "10Y", "t": 10.0, "rate": 4.08}, …],
  "compare": {"1w": {"day": "2026-10-30", "par": […]}, "1m": {"day": "2026-10-06", "par": […]}, "1y": …},
  "zero": [{"t": 0.083333, "label": "1M", "zero": 3.9725, "df": 0.99672741}, …],
  "forwards": [{"key": "1y1y", "start": 1.0, "end": 2.0, "forward": 3.3739}, …],
  "measures": {"3m10y": 20.0, "2s10s": 56.0, "5s30s": 105.0, "2s5s10s": -38.0, "be5": 2.19, "be10": 2.22},
  "conventions": {"interpolation": "…", "bootstrap": "…", "zero": "…", "forward": "…"},
  "source": "U.S. Department of the Treasury, daily Treasury par yield curve rates (public domain)"
}

POST/v1/rates/bond

A fixed-coupon bond: settlement, maturity, coupon (percent of face) and either the clean price per 100 or the yield (percent); optionally frequency (1, 2, 4 or 12; 2 by default), day_count (30/360, US, by default, or ACT/ACT, ICMA), redemption, face and calls ([{"date": "2031-11-15", "price": 100}]). The clean and dirty price, accrued interest, the yield to maturity (compounded at the coupon frequency, simple interest in the final period), Macaulay and modified duration, convexity, DV01, the yield to each call and to worst, and the G-spread and Z-spread over the Treasury curve on file on the settlement date, with the cash flows.

curl https://www.tibeb.trade/v1/rates/bond \
  -H "Authorization: Bearer $TIBEB_KEY" \
  -H "Content-Type: application/json" \
  -d '{"settlement": "2026-11-09", "maturity": "2036-11-15", "coupon": 4.25, "price": 99.5,
       "calls": [{"date": "2031-11-15", "price": 100}]}'

{
  "clean": 99.5, "dirty": 101.554167, "accrued": 2.054167, "ytm": 4.311907,
  "macaulay": 8.088608, "modified": 7.917901, "convexity": 76.075, "dv01": 0.08041,
  "calls": [{"date": "2031-11-15", "price": 100.0, "ytc": 4.361882}], "ytw": 4.311907, "ytw_is": "maturity",
  "g_spread_bp": 23.06, "z_spread_bp": 23.35, "curve_day": "2026-11-06",
  "cash_flows": [{"date": "2026-11-15", "t": 0.016667, "amount": 2.125}, …],
  "conventions": {"day_count": "…", "yield": "…", "dv01": "…", "g_spread": "…", "z_spread": "…"}, "missing": []
}

POST/v1/rates/mm

A bill on each basis: days to maturity (or settlement and maturity) and exactly one of price (per 100), discount, bey or mmy (percent). The price, the bank-discount yield and the money-market yield (both ACT/360) and the bond-equivalent yield — the Treasury's investment rate: simple interest out to half a year, its quadratic beyond, on a 366-day year when the twelve months after settlement hold a 29 February.

curl https://www.tibeb.trade/v1/rates/mm \
  -H "Authorization: Bearer $TIBEB_KEY" \
  -H "Content-Type: application/json" \
  -d '{"days": 91, "discount": 3.79}'

{"days": 91, "given": "discount", "price": 99.041972, "discount": 3.79, "mmy": 3.82666, "bey": 3.879809,
 "year_days": 365, "conventions": {…}}

GET/v1/agents

The agents there are, their steps, and which steps wait for approval.

POST/v1/agents/{agent}/runs

Start an agent on a thesis. start_at (optional) is the step to start from.

curl -X POST https://www.tibeb.trade/v1/agents/committee/runs \
  -H "Authorization: Bearer $TIBEB_KEY" \
  -H "Content-Type: application/json" \
  -d '{"thesis": "w1"}'

{
  "id": "run_3f9a0c1b2d4e5f60",
  "agent": "committee",
  "thesis": "w1",
  "status": "queued",
  "steps": [
    {"n": 0, "name": "committee", "status": "queued", "started": null, "finished": null,
     "seconds": null, "cost_usd": null, "words": "", "error": "", "gate": null,
     "needs_approval": false}
  ],
  "created": "2026-11-09T09:00:00+00:00",
  "updated": "2026-11-09T09:00:00+00:00",
  "cost_usd": 0,
  "input": {"thesis": "w1", "engine_id": "t_…", "start_at": 0, "auto": false}
}

A run's status is queued, running, waiting, completed, failed, cancelled or interrupted (the server restarted under it). A failed step says why in plain words and skips the steps after it. A run never carries a login id or an email.

GET/v1/runs

Your runs, newest first.

GET/v1/runs/{id}

One run, as above.

GET/v1/runs/{id}/events

Server-sent events: an event: run with the whole run each time it changes, then event: end when it has ended.

curl -N https://www.tibeb.trade/v1/runs/run_3f9a0c1b2d4e5f60/events \
  -H "Authorization: Bearer $TIBEB_KEY"

POST/v1/runs/{id}/approve

Approve a waiting run. A step that asked a question takes an answer: {"answer": {"choice": "…", "fields": {…}, "url": "…"}}; no body approves. 409 when the run isn't waiting.

POST/v1/runs/{id}/cancel

Stop a run between steps. 409 once it has ended.

POST/v1/runs/{id}/steps/{n}/rerun

A new run of the same agent on the same thesis, from step n on.

POST/v1/webhooks

{"url": "https://…", "events": ["run.completed"]} — events is optional (all three). The address must be https, with no credentials in it, and on the public internet: a private, loopback, link-local or carrier-grade NAT address is refused when it is registered and again each time a hook is sent, and the request goes to the very address that was checked.

{
  "id": "hook_8c1d2e3f4a5b",
  "url": "https://example.com/tibeb",
  "events": ["run.completed", "run.failed", "run.waiting"],
  "secret": "whsec_…",
  "note": "Keep the secret: it is not shown again."
}

GET/v1/webhooks

Your webhooks (without their secrets) and the events there are.

DELETE/v1/webhooks/{id}

Remove one.

POST/mcp

The Model Context Protocol over HTTP (JSON-RPC 2.0, protocol 2025-06-18), with the same key. Methods: initialize, ping, tools/list and tools/call; a notification (no id) gets 202 and no body, and a batch (a JSON array) gets an array back. The tools are the REST routes under other names, so your code and your AI tools see one thing.

GET/v1/plans the owner's key only

The desk's plans.

GET/v1/journal the owner's key only

The desk's journal and its scorecard. Any other key gets 403.