Home / Agents

🤖 641-agent library · public API

641 specialist agents. One API.

DCS Agents is the orchestrated fleet that powers every Platform build — 48 page-producing, 590 industry sub-agents, 37 support-pipeline, 9 shared, 1 orchestrator. Available as a public API, an MCP server and a visual studio — each agent carries an on-chain Reputation SBT.

Get an API key Open Agent Studio

The 641-agent fleet

Composition, named honestly

Not one giant model. A fleet of specialists, orchestrated.

641specialist agents · orchestrated for every Platform build

Page-producing

Architect, PM, Designer, Frontend, Multi-Page Router, SEO, Accessibility, Performance, Mobile, Security, QA+Testing, Output Validator, DevOps + 35 more.

590

Industry sub-agents

Per-vertical specialists across the 22 launch verticals — restaurant menu writer, clinic intake form drafter, salon-services drafter, etc.

Support pipeline

Image generation, copy editing, asset compression, sitemap builder, schema markup, OG cards, favicon, lighthouse fix.

Shared

Cross-vertical agents — brand-tone setter, accessibility check, contrast guard, i18n string extraction, sentinel, build-quality score.

Orchestrator

The conductor — reads the brief, picks the right specialists, queues parallelism, watches the build-quality score, decides when the site is shippable.

22 launch verticals

One catalog. Twenty-two verticals.

Each vertical bundles its industry sub-agents, templates and memory schema.

🍽Restaurantv22.r1

☕Cafév22.r2

🩺Healthcarev22.h1

🦷Dentalv22.h2

🏠Real Estatev22.re

🛒E-commercev22.ec

👗Fashionv22.fa

💄Beauty & Salonv22.bs

🏋Fitnessv22.fi

⚖Law Firmv22.lw

📊Accountingv22.ac

✈Travelv22.tr

🏨Hotelsv22.ho

🚗Car Rentalv22.cr

📢Marketing Agencyv22.ma

☁SaaSv22.ss

🎓Educationv22.ed

🏗Constructionv22.co

📷Photographyv22.ph

🎉Eventsv22.ev

💼Consultingv22.cn

🔧Automotivev22.au

87 industries covered today, expanding to 107. Each vertical id is stable — pin a vertical, lock the agents.

Agent identity

Every agent has a verifiable on-chain identity

F1 Reputation SBT is live on Base mainnet — first mint executed. Operator reputation anchored on-chain alongside the receipt chain.

F1 SBT · BASE MAINNET · LIVE

Reputation SBT

Soulbound — non-transferable per-agent reputation. Mints accumulate as agents accumulate signed work; tiers (Probation → Verified → Trusted) gate capability access.

0xbDd1f5fC349D9a8EfCEb07Edbd491233b2540f5F

contract on Base · verifiable on Basescan

Why a soulbound token?
Reputation can't be bought or sold — only earned. A trusted agent is forever the agent that did the work; rep doesn't transfer if it changes hands.

What it gates.
Higher tiers unlock larger base rates on Compute, lower retry budgets, premium-vertical dispatch, and the Trust SKU surface.

Verifiable, not "trustworthy because we say so."
The contract is on Base; every mint is queryable; the receipt chain ties each agent run to the corresponding mint at verify.dcsai.ai.

Agent Studio · F1 live

Visual builder, signed output.

Compose an agent from prompt, tools and a vertical anchor; test-dispatch from the panel; deploy to your API key. Every run emits a signed R+2 receipt and contributes to that agent's SBT-anchored reputation.

Drag-and-drop prompt · tool · vertical composition
Test-dispatch panel calls the Public Agents API live
Each saved agent gets an SBT identity
Saves to POST /api/agents — rev-controlled by the chain head

Open agent-studio.html →

studio.compose()visual · F1 LIVE

// composing a new agent in the Studio UI
{
  name:      "weekend-brunch-promo",
  vertical:  "restaurant",
  prompt:    "draft a weekend brunch promo",
  tools:     ["os.broadcast", "platform.menu"],
  approval:  "required-above-100-contacts",
}

saved · SBT minted · ready to test-dispatch

F11 Public Agents API · live

The 22-vertical catalog over a stable HTTPS API.

One catalog endpoint, one run endpoint, one signed receipt chain. The same code that powers Platform builds is callable from any stack — with per-key rate limits, LRA run semantics and a public verifier.

GET /api/v1/agents — the full 22-vertical catalog
POST /api/v1/runs — create a long-running activity
API-key auth · per-key concurrency caps · receipts on every run
SDK in TypeScript and a Python client

curl /api/v1/runsF11 LIVE · api.dcsai.ai

# start a long-running activity against a vertical
curl -X POST https://api.dcsai.ai/api/v1/runs \
  -H "Authorization: Bearer $DCS_KEY" \
  -d '{
        "vertical": "restaurant",
        "agent":    "menu-writer",
        "input":    {"site_id":"bella-cucina"}
      }'

# → { run_id:"...", status:"queued", receipt:"r2:..." }

queued · receipt r2:9f3c…a17 signed

MCP server · @dcsplatform/mcp-server

Drop the full 641-agent catalog into any MCP client.

One npx command and the entire 22-vertical catalog — plus the R+2 / R+3 / R+4 verifier and the Sovereign Memory Graph — show up as tools inside Claude Desktop, Cursor, Cline or any MCP-aware client.

22-vertical catalog over MCP, with the right schema per tool
Verifier tools (R+2 / R+3 / R+4) exposed alongside
Memory graph as a stdio MCP server
Open spec at github.com/DCS-LabsAI

claude-desktop.config.jsonMCP · stdio

{
  "mcpServers": {
    "dcs": {
      "command": "npx",
      "args": ["-y", "@dcsplatform/mcp-server"]
    }
  }
}

22 verticals · verifier · memory — live as MCP tools

Shipped & coming

How is "641 agents" different from a single big LLM?

It's an orchestrated fleet of specialists, not a monolithic prompt. 48 page-producing agents, 590 industry sub-agents, 37 support-pipeline workers, 9 shared agents, 1 orchestrator. Each is a focused worker with a stable contract.

What does an Agent SBT actually do?

It anchors per-agent reputation on Base mainnet — non-transferable, accumulated through signed work. Higher tiers (Probation → Verified → Trusted) gate larger Compute base rates and premium-vertical dispatch.

Is the Public Agents API rate-limited?

Yes — per-key concurrency caps and per-day quotas tuned by plan. The catalog endpoint is unmetered; POST /api/v1/runs counts against the plan.

How does LRA work?

Create a run with POST /api/v1/runs; receive a run_id. Poll or subscribe for state transitions (queued → running → done). Each transition emits a signed receipt you can verify at verify.dcsai.ai.

Can I bring my own model?

Yes — through Sovereign. The whole fleet runs inside your VPC with your weights and your keys; the contract is the same and the receipt chain is local.

What's the latency?

For a single-agent run, sub-second to a few seconds depending on the vertical and tool calls. A full Platform build is p50 127 s today (target ≤ 90 s by Q3 2026).

Is there a Python client?

Yes — pip install dcs-agents ships the same code path as the TypeScript SDK and the MCP server.

How do I move from prototype to production?

Compose in Agent Studio → save (auto-mints an SBT) → test-dispatch from the panel → promote with POST /api/v1/agents/:id/promote. Promotion is signed and reversible.

641 specialist agents. One API.

Composition, named honestly

One catalog. Twenty-two verticals.

Every agent has a verifiable on-chain identity

Reputation SBT

Visual builder, signed output.

The 22-vertical catalog over a stable HTTPS API.

Drop the full 641-agent catalog into any MCP client.

Live in production — or arriving soon

22-vertical catalog

LRA run lifecycle

F1 Reputation SBT

F1 Multi-Modal Agent Studio

MCP server

Multi-Agent Engine (MAE)

Self-Evolving agents (F4)

A2A & Economy v1

Per-agent benchmarks API

Questions developers ask first

What sits next to Agents

Platform

Memory

Inference

DCS Labs

Pick an agent. Run it. Verify the receipt.