Run Multiple AI Agents in Parallel — 10K in One Call

Sequential agent execution is the bottleneck you've accepted as normal. Why run 1,000 scoring jobs one by one when they're independent? Slopshop Army fans out any task to thousands of parallel agents, collects results, verifies them with a Merkle tree, and returns a single aggregated response. No orchestration infrastructure required.

The parallelism gap

Most agent frameworks are built for sequential reasoning: one agent, one task, one result. But many real problems are embarrassingly parallel:

• Scoring 5,000 product descriptions for quality
• Running 1,000 Monte Carlo trials for risk modeling
• Enriching a batch of 2,000 records with entity extraction
• Generating N candidate solutions and taking the majority vote
• Testing a hypothesis across many different starting conditions

Doing any of this yourself means spinning up async workers, managing concurrency, handling partial failures, and aggregating results. That's days of infrastructure work before you've written a single line of actual agent logic.

POST /v1/army/deploy

Specify a task, a count, an optional input list, and a strategy. Army fans out the work, runs agents in parallel, and returns aggregated results with a Merkle root you can use to verify the full result set without downloading every individual output.

curl -X POST https://slopshop.gg/v1/army/deploy \
  -H "Authorization: Bearer demo_key_slopshop" \
  -H "Content-Type: application/json" \
  -d '{
    "task": "score_quality",
    "count": 100,
    "inputs": [
      {"id": "prod_001", "text": "Ultra-soft 100% cotton t-shirt with reinforced seams"},
      {"id": "prod_002", "text": "T shirt"},
      {"id": "prod_003", "text": "Premium moisture-wicking athletic tee, 4-way stretch fabric"}
    ],
    "strategy": "map",
    "params": {
      "criteria": ["clarity", "completeness", "tone"],
      "output_format": "score_0_to_10"
    }
  }'

{
  "ok": true,
  "army_id": "army_4d8f2a",
  "agents_deployed": 100,
  "strategy": "map",
  "results": [
    { "id": "prod_001", "score": 8.4, "breakdown": { "clarity": 9, "completeness": 8, "tone": 8 } },
    { "id": "prod_002", "score": 1.2, "breakdown": { "clarity": 2, "completeness": 1, "tone": 1 } },
    { "id": "prod_003", "score": 9.1, "breakdown": { "clarity": 9, "completeness": 10, "tone": 9 } }
  ],
  "merkle_root": "a3f9c2d8e1b047...",
  "total_time_ms": 412,
  "_engine": "real"
}

Monte Carlo simulations

Set strategy: "monte_carlo" and Army runs N independent trials with randomized parameters, then returns the distribution of outcomes. Use this for risk modeling, sensitivity analysis, or any problem where you want to understand variance rather than a single answer.

curl -X POST https://slopshop.gg/v1/army/deploy \
  -H "Authorization: Bearer demo_key_slopshop" \
  -d '{
    "task": "revenue_projection",
    "count": 5000,
    "strategy": "monte_carlo",
    "params": {
      "base_price": 49.99,
      "price_variance": 0.15,
      "demand_elasticity": -1.2,
      "market_size": 50000,
      "market_variance": 0.25
    }
  }'

# Returns: distribution stats (p5, p25, p50, p75, p95),
# histogram data, and full trial results with merkle_root

{
  "ok": true,
  "strategy": "monte_carlo",
  "trials": 5000,
  "distribution": {
    "p5": 1820400,
    "p25": 2104000,
    "p50": 2389500,
    "p75": 2741000,
    "p95": 3218000
  },
  "mean": 2402100,
  "std_dev": 421800,
  "merkle_root": "b7e3a1f0c9d2...",
  "_engine": "real"
}

Strategies

Army supports four execution strategies. Pick the one that matches your problem:

Merkle verification

Every Army response includes a merkle_root. This is the root of a Merkle tree built from hashes of all individual agent results. You can use it to:

• Verify that a specific result was part of the run, without downloading the full result set
• Prove to a third party that the aggregated output hasn't been tampered with
• Cache only the Merkle root and fetch individual proofs on demand
• Audit that 1,000 agents actually ran rather than one agent pretending to run many times

curl -X POST https://slopshop.gg/v1/army/army_4d8f2a/verify \
  -H "Authorization: Bearer demo_key_slopshop" \
  -d '{
    "result_id": "prod_001",
    "merkle_root": "a3f9c2d8e1b047..."
  }'

# Returns: { "included": true, "proof": [...], "leaf_hash": "..." }

Scale limits

Pricing

Army charges per agent deployed, not per call. Standard task agents cost 2 credits each. Monte Carlo trials cost 1 credit each (lighter compute). You get 500 free credits on signup — enough for 1,000 free agent runs to validate your use case. See full pricing.