Live · loading… →

Make your product instantly callable by every AI agent.

Users won't read your docs anymore — their AI agents will. DaaS turns your product documentation into a structured skill API that agents route, understand, and act on. Higher activation. Lower support load. Zero manual maintenance.

Open Console → How it works
Works with
Apache-2.0 · self-hosted · any LLM provider · multi-tenant
Powered by · Default backed by Qwen · swap to Claude / GPT / any provider in one config line
DaaS Pipeline 
$ daas generate -c products/my-product.json
  ✓ Phase 0: discover     · 1,847 docs in 3 dirs
  ✓ Phase 1: extract      · 1,847/1,847 (qwen-flash, parallel)
  ✓ Phase 2: organize     · 23 capability domains
  ✓ Phase 2c: intents     · 6 intent skills
  ✓ Phase 3: detail       · 41 SKILL.md files
  ✓ Phase 3b: intent      · adversarial recall 96%
  ✓ Phase 4: entry        · skills/my-product/SKILL.md

$ curl https://your.daas/api/route?q=how+do+i+deploy
[
  { "skill": "my-product-deploy-model",
    "score": 285,
    "reason": "intent + 3-token overlap" }
]▍
—
Products live
—
Intent skills
—
Detail skills
96%
Adversarial recall
~10s
Cache-hit rebuild
Knowledge Explorer

Explore the live skill tree.

Every product below was auto-generated by DaaS — crawled, organized, and validated. Click any product to browse its skill breakdown.

GET /api/products/oss
Skill Spotlight

See what a generated skill looks like.

Each skill includes decision trees, comparison tables, and actionable steps — all auto-generated from raw documentation.

GET /api/skills/oss/intent/oss-manage-objects
Routing Playground

Try the routing engine live.

Ask anything — DaaS instantly ranks the best-matching skill across every product. Same engine your agents call in production.

How it works

From raw docs to an agent surface in minutes.

Point DaaS at your documentation — any URL, any format. Within minutes your product is callable by every AI agent: Claude, GPT, in-house copilots. No manual tagging, no embeddings to retrain, no maintenance.

01

Crawl & Understand

Give a URL — DaaS crawls every page, understands its purpose, and extracts structured facts (endpoints, parameters, prerequisites, error codes).

parallel · checkpoint-resumable · any format
→
02

Map User Intents

Docs are grouped by capability, then cross-referenced to discover what users actually want — "How do I deploy?" maps to 4 valid paths with trade-offs.

auto-consolidation · multi-path discovery
→
03

Generate & Validate

Each skill goes through a 6-step pipeline with 6 quality validators. Decision trees, comparison tables, FAQs — all verified before shipping.

adversarial recall > 96% · critique loop
→
04

Ship Agent-Native

Your product is now an HTTP API any agent can route to. User asks → agent finds the right skill → action happens. Sub-millisecond routing, zero hallucination.

HTTP API · SDK · MCP-ready · multi-tenant
RAG vs DaaS

Why structured skills beat chunked retrieval.

Traditional RAG splits your docs into chunks and hopes the right ones surface. DaaS understands intent first, then routes to deterministic, validated skill content.

R

Traditional RAG

Chunk #47 (cosine 0.81)
...the model can be deployed using EAS. Before deployment, ensure the resource group has been created. For GPU instances, check the quota in...
Partial
Chunk #112 (cosine 0.78)
...OSS provides object storage for unstructured data. Buckets can be configured with lifecycle rules, versioning, and cross-region replication...
Irrelevant
Chunk #203 (cosine 0.74)
...for distributing models via OSS, first upload the model artifact to an OSS bucket, then reference the OSS path in the EAS service config...
Relevant
vs
D

DaaS Skill Routing

Loading live result...
Routing Latency
~2.3s <1ms
Embedding Tokens
~4,200 0
Accuracy
Partial Deterministic
Architecture

A 7-phase LLM pipeline under the hood.

Every judgment call — what's a domain, what's an intent, which keywords match a query — is delegated to an LLM. Python orchestrates parallelism, 4-layer caching, and 7 deterministic validators.

0
Discover
scan dirs
1
Extract
JSON facts
2
Organize
domains
2b
Paths
alternatives
2c
Intents
user goals
3
Generate
validate
4
Ship
entry

Three consumption surfaces

Direct file load

Claude Code, Cursor, or any agent reads SKILL.md from disk

skills/{product}/SKILL.md

HTTP API + Web UI

Single or multi-tenant, with telemetry and workspace isolation

GET /api/route?q=...

pip install daas-sdk

LocalSkillRouter or RemoteSkillRouter for Python agents

router.route(query)

7 Deterministic Validators

ValidatorCatches
decision_tree_falsifiabilitySubjective conditions
faq_counter_examplesMissing wrong-path entries
faq_dedup_with_decision_treeFAQs that rephrase the tree
wins_loses_complementarityAsymmetric trade-offs
keyword_coverageMissing technical terms
reverse_validationContradicted linked skills
path_label_honestyUndocumented path verbs

Cost & Latency (2,000 docs)

PhaseColdCachedCost
Extract~25 min~5 s$2.00
Organize~3 min<1 s$0.50
Paths + Intents~2 min<1 s$0.40
Generate~3 min<1 s$2.00
Intent skills~5 min<1 s$5.00
Total~37 min~10 s~$10
4-layer smart cache
CompressionSHA256(chunk) ExtractionSHA256(content+prompt) OrganizationSHA256(manifest) PathsSHA256(summary)
Built in the open

No fake logos. Just the numbers, the stack, and the source.

This page renders from a single process on a 1-vCPU box behind Cloudflare. Every counter below is live; every line of the engine is on GitHub.

📡 Live from this server
  • —products generated
  • —SKILL.md files served
  • 96%adversarial recall

Counters pulled from /api/dashboard/summary. Click any of the stat-bands above to verify.

🧱 Stack & provenance
  • Skill format follows Anthropic's anthropics/skills marketplace spec verbatim
  • Default LLM is Qwen via DashScope; swap to Claude, GPT or vLLM in one config line
  • Apache-2.0 engine + BSL multi-tenant ops. Self-host single-tenant is free forever
  • Zero third-party Python dependencies at runtime — `pip install` then `python3 -m`
💸 Honest costs
  • $1.50cold-start, 13k docs, qwen-flash
  • $0.05/daysteady-state across 16 products
  • 50rpssustained on 1 vCPU + 2 GB

Numbers from the reference deployment running this site. Not extrapolated.

One-click distribute

Ship every workspace as a Claude Code plugin.

Hit Publish in your dashboard. DaaS pushes the workspace to a GitHub repo formatted exactly like Anthropic's official anthropics/skills marketplace. Users install your skills with one slash command.

1
Click Publish in DaaS
Per workspace · creates the repo on first push
→
2
DaaS commits to GitHub
daas-skills-{workspace} · marketplace.json + README + skills/
→
3
User installs in Claude Code
One command. Skills load on demand.
Claude Code 
> /plugin marketplace add chenkai66/daas-skills-vercel
  ✓ Marketplace 'daas-skills-vercel' added

> /plugin install vercel@daas-skills-vercel
  ✓ Installed plugin 'vercel' · 28 skills loaded

> How do I deploy a Next.js app to Vercel?
  → routing to vercel-deploy-application (intent)
  → loading vercel-deployment (guide)
  Step 1: Connect your Git provider in Vercel Dashboard...▍
Anthropic-spec compliant Same .claude-plugin/marketplace.json format the official skills repo uses.
Auto-discovered Aggregator sites (claudemarketplaces.com, lobehub) crawl GitHub — your repo shows up automatically.
Versioned + reproducible Each publish is a UTC-timestamped commit. Re-publish on doc updates without breaking installs.
Knowledge Graph

See how your products connect.

Every product, intent, and cross-product combo mapped as a live force-directed graph. Edges represent real co-occurrence in routing telemetry — not hand-drawn diagrams.

—Products
—Cross-product edges
—Discovered combos
Generative Engine Optimization

Get cited inside the answer, not just ranked.

Classic SEO ranks you on Google. GEO gets your product recommended by ChatGPT, Perplexity, Claude and Gemini. DaaS already structures your docs into skills — GEO exposes them on every surface a generative engine crawls, parses, and quotes, then measures whether you actually surface.

—
Platform GEO score
—
Products optimized
—
Citable AI surfaces
Welcomed crawlers GPTBot ClaudeBot PerplexityBot Google-Extended Gemini
01

Generate test queries

Brand-free questions are synthesised straight from each intent — so we measure organic surfacing, never a brand lookup.

from the skill graph · zero hand-written prompts
→
02

Measure & diagnose

Each query runs through our own router. A miss is split into coverage gap vs surfacing gap vs wrong product — the exact thing to fix.

/api/route is the target AND the ruler
→
03

Guarded rewrite

GEO-paper levers (answer-first, statistics, citations) are applied to the weakest skills — guarded by a key-point-recall check so accuracy is never traded for visibility.

no facts, endpoints or numbers dropped
Explore GEO scores → Run the loop in console →
Agent Integration

Integrate in under 5 minutes.

One HTTP call. Structured JSON back. Works with any language, any agent framework. Here is a live example calling the DaaS routing API.

Terminal 
curl -X POST https://company-skill.com/api/route \
  -H "Content-Type: application/json" \
  -d '{"query": "how do I deploy a model on PAI", "top_k": 3}'
from daas_sdk import RemoteSkillRouter

router = RemoteSkillRouter(base_url="https://company-skill.com")
matches = router.route("how do I deploy a model on PAI", top_k=3)

for m in matches:
    print(f"{m.skill_id} (score: {m.score})")
    print(router.load_match(m)[:200])
# In Claude Code, install the DaaS skills plugin:
/plugin marketplace add your-org/daas-skills-pai

# Then just ask naturally:
> How do I deploy a model on PAI?
  → routing to pai-deploy-model (intent)
  → loading pai-deployment (guide)
  Step 1: Create an EAS service...
Features

Enterprise-ready, built for scale.

Live workspace

Already powering — products live.

From Alibaba Cloud infrastructure to international SaaS platforms — each chip is a real product whose docs DaaS has ingested, organized, and made query-ready.

Browse all in console →
ROI Calculator

See how much your team saves with DaaS.

Plug in your team's numbers — see the cost of context-switching to documentation that DaaS eliminates.

Hours saved/week per dev
—
Annual savings (whole team)
—
DaaS Pro annual cost
¥3,588
ROI multiple
—×
Assumptions
  • Avg developer touches 6 SaaS / cloud APIs per week with non-trivial tasks.
  • Each unfamiliar task: ~25 min context-switching + reading docs.
  • With DaaS, the AI agent routes directly to a validated skill: ~3 min.
  • Saved per task: 22 min × 6 tasks = 2.2 hr per dev per week.
  • 52 weeks/year, salary spread across 2,000 productive hours.
Pricing

A marketplace for product intelligence.

Product companies publish their docs as AI-callable skills. Agent developers query them all through a single API.

Team
¥999 / mo · $149
1 workspace · 5 products · 100k routes/mo
  • Hosted skill generation from your docs
  • Weekly auto-rebuild on doc change
  • Distribution to all MCP clients
  • Email support · best-effort SLA
Scale
Business
¥4,999 / mo · $699
5 workspaces · unlimited products · 1M routes/mo
  • Everything in Team
  • Daily auto-rebuild · 99.5% SLA
  • Slack/Lark support · 24h response
  • Telemetry dashboard + Gap Radar
Contact Sales
Community
Free · Apache-2.0
Self-host single-tenant · no limits
  • Full pipeline source on GitHub
  • Python SDK + per-product MCP servers
  • All 7 deterministic validators
  • Multi-LLM (DashScope / OpenAI / Anthropic / vLLM)
Get Started
Hosted
Team
¥999 / mo · $149
100k routes/mo · zero ops · 14-day trial
  • Hosted multi-tenant routing
  • Weekly auto-rebuild on doc change
  • Telemetry + Doc Gap Radar (LLM-clustered)
  • Overage: ¥5 / 1k routes (or $0.75)
🧪 Demo mode is on — judges can test the full Subscribe → QR → auto-pay → API key flow without spending money.

Make your product agent-native today.

Every product needs an agent surface — or it gets routed around. DaaS gives yours one in minutes. Apache-2.0 open source, or fully managed.

Open Console →