The Agentic Workflow Explorer

Run Any of Finmagine's 10 Pre-Built Agent Recipes End-to-End — Zero Code, Live DAG, Real Results

COMPLETE GUIDE

Building with the API? Open the developer docs

Finmagine API & MCP — free tier to test • Pro/Max for full access • REST + MCP

Open the Docs →
← API & Agent Integration Hub ← All Feature Guides
Published: July 29, 2026  |  9 min read  |  Developer Guide  |  API & Agent Integration Series

Multimedia Learning Hub

Master the Agentic Workflow Explorer through a complete learning overview and interactive flashcards

What You Will Master

The Interactive API Explorer is for trying one operation at a time. The Agentic Workflow Explorer is its counterpart for multi-step workflows — it runs the actual recipes published in agent-recipes.json, chaining each step's real response into the next, so you can watch a full agent workflow execute before writing any orchestration code yourself.

What This Guide Covers:

  1. The 10 Recipes at a Glance — what's visible by default vs. behind the LLM-required toggle
  2. Anatomy of a Recipe Run — the live DAG, step cards, and provider/write warnings
  3. Honest DAG Connectors — why some steps show a real dependency label instead of a simple arrow
  4. Two Ways to Run — Run All Steps vs. Run Next Step, and the confirm-to-arm gate before real spend
  5. Async Recipes: Polling, Cancel, and the Pending-Jobs Banner
  6. Copy Workflow Instructions — the escape hatch to your own code or an agent prompt
  7. Which Recipe to Try First — a practical starting order

Who This Is For:

  • Developers evaluating a multi-tool workflow before coding it — see the real chained responses first
  • Agent builders — understand exactly how depends_on/input_map chaining works by watching it happen
  • Anyone about to spend real Investment Committee or Research Preset quota — understand the confirm-to-arm gate and Cancel button before you click Run

Test Your Knowledge

Click any card to reveal the answer. Use the search box to find a specific topic.

The 10 Recipes at a Glance

Every recipe the Explorer can run comes straight from agent-recipes.json — nothing is hand-typed into the page separately. 6 need no LLM key and are visible immediately; 4 require a real LLM provider and stay hidden behind a toggle, off by default.

Visible by default — no LLM key required

RecipeMarketTierSteps
BYO LLM Key Setup—Free3 (prerequisite for the LLM recipes below)
India: Zero-Cost Technical DemoIndiaAPI Pro2
US: Zero-Cost Technical DemoUSAPI Pro2
India: Momentum SetupIndiaAPI Pro4
US: Momentum Setup (partial)USFree3 (deliberately partial)
India: Post-Earnings ReviewIndiaAPI Pro4

Behind the "Show LLM-required workflows" toggle

RecipeMarketTierAsyncSteps
India: Screen → Investment CommitteeIndiaAPI MaxYes7
India: Trader Scan → Investment CommitteeIndiaAPI ProYes5
US: Trader Scan → Investment CommitteeUSAPI ProYes5
Research Preset Deep Dive—API ProYes4
Why is BYO LLM Key Setup always visible even though one of its steps touches a real provider? It's the prerequisite recipe — it's how you save and verify the provider key everything else needs, so hiding it behind the same toggle it unlocks would be circular. Its one provider-touching step (verifying the saved key) is clearly warned on its own step card instead.
STEP 1

Anatomy of a Recipe Run

Selecting a recipe renders it as a horizontal row of step cards — a simplified DAG (directed acyclic graph) view. Each card shows:

Clicking any step card opens its full detail — the exact request that will be sent, and once run, the exact response received.

Provider check comes before the general "writes to your account" warning, on purpose. A step like test_byo_key is technically a write-shaped call (Account+POST) but doesn't actually change anything in your account — it only verifies a saved key. Labeling it "writes to your real account" would be misleading; the more relevant warning is that it calls a real provider.

Honest DAG Connectors

Most steps in most recipes really are a simple chain — each one depends only on the step immediately before it, so a plain → arrow between cards is accurate. But not always, and the Explorer doesn't pretend otherwise.

// India: Screen → Investment Committee has a step called // allocate_committee_risk that genuinely has THREE parents — // it needs data from screen, report, AND risk_compute, not just // whichever card happens to sit to its left in the layout. screen ──→ fast_scan ──→ start ──→ poll ──→ report ──┐ ├──⤷ allocate_committee_risk risk_compute ──┘

Whenever a step's real dependencies don't match "the card immediately to its left," the connector switches from a plain arrow to a merge symbol with an explicit label — ⤷ from screen, report, risk_compute — naming every real parent instead of silently drawing a misleading straight line.

This was a real bug caught by external review, not a feature planned from day one. An earlier version of the Explorer laid every recipe out as a simple left-to-right pipeline. A reviewer flagged that this became actively misleading the moment a genuinely multi-parent step existed — the fix shown above is a small, proportionate one (a text label) rather than a full auto-layout graph engine, since even the most complex recipe today has at most three parents feeding one step.
STEP 2

Two Ways to Run

Every recipe offers two run modes, plus two secondary utilities:

ButtonWhat it does
▶ Run All StepsExecutes every step in dependency order automatically, chaining each response into the next, stopping only on an error or a confirm-to-arm gate
⏭ Run Next StepExecutes just the single next runnable step (the first one whose dependencies are all done), then waits for you to click again
↺ ResetClears all step statuses and results, letting you re-run the recipe from scratch
📋 Copy Workflow InstructionsExports the full tool-call sequence as plain text — see the dedicated section below

The confirm-to-arm gate

For any recipe that touches a real LLM provider or spends shared quota, neither run button fires on the first click. The first click arms the action and shows exactly what's about to happen — which provider, and for a preset run, the actual preset and goal that will be sent — and only the second click actually executes it. Run Next Step has its own, narrower version of this gate scoped to just the specific step about to run, separate from Run All Steps' recipe-wide gate.

Before either run button unlocks at all on an LLM-required recipe, you need a saved provider key selected. A key panel appears above the run buttons — "Select a saved provider key for this workflow's LLM step" — and both Run All Steps and Run Next Step stay disabled until you've picked one.
STEP 3

Async Recipes: Polling, Cancel, and the Pending-Jobs Banner

The 4 Investment-Committee and Research-Preset recipes are async — starting the job is instant, but getting the finished result takes real time. The Explorer handles this with a dedicated poll step rather than making you refresh manually.

Differentiated poll timeouts

Poll toolMax wait before giving up
check_committee_status~12 minutes
check_preset_status~60 minutes

These caps are deliberately different — Research Presets can genuinely run 15-40 minutes depending on which of Strategy Lab's 8 analyst chains you picked, so a single one-size-fits-all timeout tuned for the faster Committee runs would incorrectly mark a still-running Preset job as failed.

While a poll step runs, its card shows a live sub-state line — Polling… attempt 6, 36s elapsed — status: running — updating roughly every 6 seconds, plus a ⏹ Cancel button.

Cancel doesn't just stop watching — it tells the backend to actually cancel the job. Clicking it calls the real cancel_committee or cancel_preset operation (the same one you'd call yourself), and the actual response from that call is included in the run transcript — not a placeholder — so you can confirm the cancellation genuinely went through server-side.

What happens on timeout — and why it's different from cancel

If a poll step exhausts its max attempts without reaching a terminal status, the Explorer marks it as an error locally — but deliberately does not clear it from the pending-jobs registry the way a real terminal outcome (completed, failed, cancelled, or an explicit user cancel) does. The job may genuinely still be running server-side; giving up watching in this browser tab shouldn't make the Explorer forget a job that's still spending real quota.

The pending-jobs banner at the top of the page exists for exactly this reason. If you navigate away or refresh mid-poll, any job still registered there stays visible on your return, so a long-running Research Preset you kicked off doesn't just silently disappear from view.

Copy Workflow Instructions

Once you've seen a recipe work end-to-end in the Explorer, the 📋 Copy Workflow Instructions button exports it as plain text — ready to paste into your own code, a README, or directly into an AI agent's prompt:

# India: Screen → Investment Committee The flagship workflow: filter the full NSE universe by fundamental/ technical criteria, free-scan the shortlist, then run a full multi-lens LLM committee analysis, risk audit, and position sizing on the survivors. Market: india Required tier: api_max Quota: 5 committee runs / 24h Needs a saved BYO provider key — call list_byo_keys first if unsure which to pass. Steps (in order): 1. screen_stocks — GET /api/v1/screener Filter by ROCE/PE/market cap/RS rating/sector etc. Returns companies[] with symbol. 2. fast_scan_committee — GET /api/v1/committee/fast-scan Free, zero-LLM deterministic pre-screen across 3 lenses. depends on: screen ... (remaining steps) ... Auth: header X-Api-Key: <your fmk_... key> Docs: https://finmagine.com/developer/docs.php
This is deliberately not a native browser alert or clipboard-permission popup that blocks the page. It copies silently to your clipboard and flashes the button label to "✓ Copied" — if clipboard access isn't available, it falls back to showing the text in a selectable box instead, so the export always works one way or another.

Which Recipe to Try First

Your situationStart with
Brand new to the Explorer, want to see something run in under a minuteIndia or US: Zero-Cost Technical Demo — 2 steps, no LLM key, no committee-run quota spent
Want to try the Investment Committee or Research Presets eventuallyBYO LLM Key Setup first — it's the prerequisite everything LLM-required depends on
Curious how a real multi-source RS/breadth workflow chains togetherIndia: Momentum Setup — 4 independent read endpoints, no mutation, no provider calls
Ready to see the flagship end-to-end, real committee outputIndia: Screen → Investment Committee — the 7-step, 3-parent recipe covered above
Every LLM-required recipe run here is real, not simulated. Running India: Screen → Investment Committee to completion genuinely spends one of your 5 daily committee-run quota slots and calls your chosen LLM provider — the same as triggering it from the Strategy Lab page itself. Use the Zero-Cost Demo recipes first if you just want to see the Explorer's mechanics without spending anything.

What's Next

With both zero-code tools covered — the single-operation Explorer and this multi-step Workflow Explorer — Wave 2 continues into the systems those LLM-required recipes actually depend on:

See the API & Agent Integration hub for the full 12-guide roadmap.

Ready to Analyse Indian Stocks Like a Pro?

Finmagine gives you 30+ computed financial ratios, sector benchmarks, FII/DII flows, the Finmagine Score, and AI-powered analysis — all in one place.

Analyse a Stock → Create Free Account
← API & Agent Integration Hub ← All Feature Guides
↑