Use case · Deep research

A week of diligence.
Done before lunch.

The bottleneck in research was never reading speed — it was that one person can only read one thing at a time. Swarms removes the 'one person' part.

A diligence brief has twelve threads the market, the team, the filings, the competitors, the customers, the lawsuits. A human works them one by one and the picture arrives in pieces. A swarm works them in parallel and the picture arrives at once.

9:12 AM

The brief lands.

A partner forwards the deal. Your agent doesn’t open forty tabs — it writes six research briefs and files one API call. Note the last task: one worker’s entire job is to argue against the deal. Try staffing that with humans who all want the deal to close.

one request — the whole engagement
POST /api/v1/swarms{  "objective": "Should we lead the Series B in Meridian Robotics?",  "tasks": [    "Map the warehouse-robotics market and its 3 largest players",    "Profile the founding team, prior exits, and key departures",    "Summarize all public filings, patents, and litigation",    "Collect customer sentiment from reviews and forums",    "Stress-test the pricing model against competitors",    "Draft the bear case — argue AGAINST the deal"  ],  "aggregatorTask": "Merge into an investment memo with a clear recommendation",  "budgetUsd": 4.00}

9:13 AM

Six analysts clock in.

Each task becomes a sandboxed worker with the deal context attached. They read filings, crawl reviews, and pull market data simultaneously — nobody waits for anybody.

The budget is a hard $4.00 ceiling across the whole run. If a worker rabbit-holes, it gets stopped at its share of the line — the run degrades gracefully instead of the bill exploding.

one call · four workers · one answer

you
researchrunning
draftrunning
fact-checkrunning
pricingrunning
merged

9:41 AM

One memo comes back. With receipts.

The aggregator merges six streams into one investment memo: recommendation up top, bear case given its own section, every claim traceable to the worker that produced it.

Beside it, the ledger: what ran, for how long, for how much — line by line, append-only. When the IC asks “where did this number come from?”, the answer is a job id, not a shrug.

the ledger never lies — append-only

holdreserved before the run−$3.00
charge43 GPU-seconds metered−$0.86
releaseunused hold returned+$2.14
receiptsigned, immutable, yours·

10:15 AM

The human does the human part.

By 10:15 the partner is doing the only work that ever needed a partner: judging the argument. The reading, collating, and first-drafting — the week of it — happened while the coffee was still warm.

the division of labor

swarmread 1,400 pages, 6 threads, in parallel
swarmdrafted the memo + the bear case
youchallenged the thesis
youmade the call
An analyst-week costs a few thousand dollars and arrives Friday. This ran for 28 minutes and cost $3.61 — and the bear case was the best-argued section in the memo.
The uncomfortable math

0

threads researched at once

0 min

brief to finished memo

$0.00

total, ledgered to the second

The same shape, everywhere

Anywhere the job is “read a lot, think a little,” a swarm wins.

Competitive teardowns

One worker per competitor. Every pricing page, changelog, and job posting — weekly.

Literature reviews

Forty papers, one worker each, merged into a single methods-aware synthesis.

RFP & vendor analysis

Each proposal scored against the same rubric by workers that never get tired on page 60.

Regulatory scans

New rules across nine jurisdictions, summarized with citations before the team logs on.

Expert-call prep

Everything a guest has ever said or published, briefed into ten sharp questions.

Post-mortem evidence

Logs, tickets, and threads read in parallel; a timeline with sources, not vibes.

Keep exploring

Give your agent a workforce.

Start free. Spawn your first swarm in minutes. Pay only for the GPU-seconds you use.