Your portfolio.Researched in depth.

A local-first research workspace for Indian equities. Explore your holdings, inspect the evidence, and make your own decisions.

FreemacOS 26+Apple SiliconInstall details

Interactive · sample book
Portfolio · read-onlysample book
  • SourceZerodha Kite · read-only scope, or a CSV you import
  • Holdings8 · ACME 21% · BHARAT 17% · CRANE 13% · NEXUS 12% · RIDGE 11% · ORION 9% · LOTUS 9% · VELUM 8%
  • Value₹8,88,705.00 · invested ₹8,21,192.50 · +₹67,512.50 (+8.2%)
  • Actionsno buy, sell or stage control exists on this surface
  • KeysmacOS Keychain · never relayed

No adapter exposes an order method, so there is nothing for a trade control to call.

AlgoagentResearch
ACMESample book

9 stages filed · audit passed

Verdict
Fair
Conviction
61/100
Confidence
81%

Return on invested capital: 18.4%.
Cost of capital: 12.1%.
Every figure has a stored source behind it.

receipt #41 · annual reportsource stored

Sample workspace · fictional book · Open the live demo.

Your keys. Your machine. Nine stages. One clear dossier. Every citation, a receipt.

Every citation is a receipt.

Algoagent Core is a free macOS research terminal for Indian equities. It turns a holding into a structured dossier, and before the model may cite a page the app has fetched and stored it — URL, text, timestamp, SHA-256 — so every number in a claim has to appear inside that stored text.

A receipt proves the source was fetched and the figure was grounded in its stored text. It does not make the source true.

Dossier · ACME illustrative data

Acme Industries Ltd · NSE · revision 3 · 9 of 9 stages filed

Verdict
Fair
Conviction
61/100
Confidence
81%

Install once, bill the maintenance annually. Return on invested capital ran 18.4% against a 12.1% cost of capital, and cash conversion held at 88%.

unknownDividend discount abstained — payout has not been stable for three years.

receipt #41bseindia.com

“Return on invested capital for the year was 18.4%. Cash generated from operations represented 88% of EBITDA.”

sha256 9e7ad4ee · fetched 14 Aug 2026 09:41 IST · application/pdf p.88

A fictional company and fictional figures, in the shape the app files on your own Mac.

Instructions are not enough. The evidence rules run in code.

Four checks in Python sit between the model and the sheet you read. A claim that fails one never reaches it.

claim pipeline · illustration fictional data
fetch ledger · local4/4 · rejected 1
Both claims are fictional. The first passes every check and is filed with its receipt. The second cites a real receipt whose stored text never printed that figure, so check two refuses it.
  1. 1 · Fetched

    A citation is an id into the local fetch ledger. A plausible URL the model typed does not resolve.

  2. 2 · Grounded

    Every figure is looked up inside the retrieved text the receipt stored, not just its address.

  3. 3 · Recomputed

    Growth and the margin bridge are calculated from statement rows. A disagreeing figure is refused.

  4. 4 · Unknown

    A gap is reported, never filled. An abstained stage caps confidence at 60%; no primary source caps it at 40%.

What it cannot do, by construction.

0 order endpoints
No adapter exposes an order method, so submitting, staging or relaying a trade has nothing to call. You act in your own exchange.
0 telemetry bytes
No usage data, no licence check, no crash upload. The diagnostics file you can export leaves the tool log out on purpose.
read-only holdings
Zerodha Kite on a read-only scope, or a CSV you import. No order scope is ever requested from your broker.
local keys
Your provider and broker keys are stored in the macOS Keychain and never relayed through us.

Nine stages the cursor cannot skip. Then a tenth that argues with them.

Nine sequential workers, context-isolated: no worker sees chat history, the operator profile, or another stage’s prose. Each has an explicit way to abstain.

  1. Establish what the business is, and whether the advantage holds.

    1. 01Mandate & Identity
    2. 02Industry & Position
    3. 03Moat & Durability
  2. Verify the numbers, read governance from filings, and close the pre-mortem before the verdict may open.

    1. 04Financial Quality
    2. 05Management & Governance
    3. 06Valuation
    4. 07Pre-mortem & Risks
    5. 08Catalysts & Monitorables
  3. Decide with quality, valuation, moat and confidence computed in Python. The schema has no field for an instruction.

    1. 09Synthesis & Verdict
10

The audit pass has no tools

A second model re-reads the frozen dossier against the code-computed values, with its tools removed and a fresh context. It can object; it can never re-score. Claims it cannot tie back to a receipt are dropped before you see the sheet.

Five surfaces on your Mac. Nothing on ours.

Research surfaces

Chat, Portfolio, Research, News, Settings

The sidebar replaces the title bar. Select a holding to open its dossier; IPO is a desk inside Research; ⌘K reaches any surface.

Local custody

Keys and research stay on the device

One OpenAI-compatible provider key, read-only broker keys, an optional fundamentals key — stored by macOS, never relayed.

Structural boundaries

The routes that do not exist

No order method on any adapter, every delivery channel off by default, and a diagnostics export that omits the tool log.

Download the terminal. Keep the research.

You pay your own LLM provider for usage and, if you want one, a fundamentals-data plan. Algoagent never handles those bills.

Check the fingerprint

Run shasum -a 256 AlgoagentCore-2.2.0.dmg and match it against the slip. That proves the file was not altered in transit.

Open it once by hand

Signed with an Apple Development certificate and not yet notarized, so Gatekeeper blocks the first launch: right-click the app in Finder and choose Open, once. This page will never ask you to strip the quarantine attribute — that disables Gatekeeper for every app you point it at.

Sign in, then paste a key

One Google or Apple sign-in: an identity check, not a paywall, kept in the Keychain and never re-validated, so later launches work offline. Then one provider key, and Kite read-only or a CSV.

Longer install notes live on Releases. Questions the page did not answer? contact@algoagent.in

Release slip 03 Sep 2026

Algoagent Core 2.2.0AlgoagentCore-2.2.0.dmg · 61 MB

Price
Free · no trial · no tier
Requires
macOS 26 or later
Architecture
Apple Silicon only
Sign-in
Google or Apple, once
Signing
Signed · not yet notarized
Markets
NSE · BSE
Download Algoagent Core 2.2.0

91be46564aa69cc528b7de6bc188b006c9adf9d57335a4f641a33dfd82ebe107

also on /releasesfree · 2.2.0 · current

Before you download.

Do I need an account?

For the Mac app, yes — one Google or Apple sign-in, so support can be tied to a real person. Nothing is gated behind money and the account unlocks no features. Algoagent stores your email, sign-in identifier and session timestamps; never your research, holdings or provider keys. The browser reconstruction asks for nothing.

What does it cost to run?

The app is free, with no trial, paid tier or usage limit. You pay your chosen LLM provider for usage, and optionally a fundamentals-data plan. Those bills are between you and them; Algoagent never sees them.

Why does macOS block the first launch?

The build is signed but not yet notarized by Apple, so Gatekeeper blocks its first launch by default. Verify the published SHA-256, then right-click the app in Finder and choose Open, once. The steps are beside the release slip and on Releases.

Is the browser version the real app?

It is a reconstruction of the macOS app’s surfaces in HTML, built from the app’s own design tokens, with fictional sample data labelled as such. The real app is a native Swift application talking to a local Python core on your Mac.

Research method · Release notes and brokers · Risk disclosure · contact@algoagent.in

You have the artifact, its fingerprint, its limits, and the choice.

Download for Mac

Free · 61 MB · macOS 26+ · Apple Silicon · v2.2.0 release notes