Decide first. Reason when it matters.

---

🔥 The problem: most LLM answers are money burned

AI applications send small decisions to large models: route this ticket, is this spam?, does this lead want to buy?, escalate to a human? Each one becomes an LLM call that:

• costs tokens — on every event, even the trivial ones;
• adds latency — the model writes text before anything happens;
• returns free text — your code has to parse it, and it can come back in a format you didn't expect;
• can hallucinate — a label that isn't in your list, a field that's missing, an action that doesn't exist.

Dcision is the decision layer that sits before your agent or LLM. You declare the questions once, in a schema. Every call returns typed answers with a confidence for each — validated against your schema — and the next action. The expensive reasoning only runs on the cases that actually need it.

⚖️ LLM vs Dcision for micro-decisions

Capability	LLM	Dcision
Cost / 1M decisions	$300 and up	~$24 incl. fallback*
Output	Unstructured text	JSON validated against your schema
Confidence	Not built in	A confidence for every answer
Next step	Parsed from free text	Policies: continue, block, escalate or fallback
Audit trail	Prompts and free text	Version, action and reason for every execution

_{* Illustrative example on the Growth plan, assuming 5% of decisions fall back to your own LLM — not a performance claim.}

🧰 Jev alone vs Dcision — everything the engine leaves to you

Dcision runs your decisions on Jev (TypeSafe's decision model) or Laya (open source, on your own server). The engine is the same — the difference is everything you don't have to build, maintain and monitor.

	Engine API alone	Dcision
Actions per result	You wire each result to the next system	8 destination types per option, level or threshold: fixed reply, structured JSON, LLM, agent, workflow, webhook, API request or function — signed deliveries with retries
Webhook trigger	You expose an endpoint and call the engine	One URL per decision — forms, CRMs, n8n, Make or Zapier run it without an API key
Dashboard	Questions live in your code	Every decision in one place, with immutable versions and full history
Decoupled	Decision logic mixed into the app	Change decisions without redeploying your app
Visual builder	Questions written by hand	Form and visual editors, JSON preview and a Playground
Templates	Start from a blank page	Ready-made decisions — lead qualification, support routing, spam, agent routing, RAG relevance…
Fallback per question	You read the confidence yourself	Minimum confidence per question and a reserved other option; below it, your policy acts
Monitoring	You build logs and metrics	Overview with volume, p50/p95 latency, errors, estimated cost and calibration suggestions
MCP	—	Remote MCP server for agents, plus a Claude Code skill
CLI and SDKs	—	Coming soon

🚀 Get started in 5 minutes

1. Create your account at app.dcision.io — Google or an e-mail code. You start on the free Genesis plan: 1M decisions a month.

2. Create a decision from a template (Lead Qualification, Support Routing, Spam Detection…) or paste one of the schemas in this repo into the editor. Try it in the Playground — Playground runs are free.

3. Deploy it and create an API key (API keys → New key):

export DCISION_API_KEY=dcs_live_...

4. Call it — cURL, Node.js or Python:

curl -X POST https://api.dcision.io/v1/decisions/lead-qualification \
  -H "Authorization: Bearer $DCISION_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"state": {"message": "We need pricing for 500 users and want to start next month.", "company_size": 500, "source": "website"}}'

{
  "result":     { "purchase_intent": 0.9412, "priority": "high", "route": "sales" },
  "confidence": { "purchase_intent": 0.8824, "priority": 0.69, "route": 0.85 },
  "action": "continue",
  "action_reason": { "type": "default" }
  // + decision_id, execution_id, version, composites, metrics…
}

Route on result and action — there's no text to parse. See Run a decision.

🔌 Call it from anywhere

Where	How
Any language	POST https://api.dcision.io/v1/decisions/{slug} with your API key (examples)
No-code, CRMs, forms	Turn on the decision's webhook URL — n8n, Make, Zapier, any form. Docs
AI agents	Remote MCP server at https://api.dcision.io/mcp. Docs
Claude Code	MCP + the Dcision skill:

claude mcp add --transport http dcision https://api.dcision.io/mcp \
  --header "Authorization: Bearer $DCISION_API_KEY" --scope user

mkdir -p ~/.claude/skills/dcision
curl -fsSL https://docs.dcision.io/skills/dcision/SKILL.md -o ~/.claude/skills/dcision/SKILL.md

🧩 Ready-made decisions

Each file has the Decision Schema (stateSchema, questions, policies…) and a sampleState to try in the Playground.

Decision	What it decides
Agent Routing	Choose the tool an AI agent should use and whether it needs an LLM.
Customer Service Router (intent routing)	Classify intent and complexity in one call: lookups go to code, billing/tech to specialist LLMs, complaints and hard cases to humans.
Lead Qualification	Score purchase intent, set priority and route an inbound lead.
RAG Relevance	Check whether a retrieved chunk answers the query before calling the LLM.
Resume Screening (composite scoring)	Score each skill on its own rubric and combine them with role-specific weights — Senior IC vs Engineering Manager.
Spam Detection	Estimate spam probability and decide whether to allow, review or block.
Support Routing	Pick the department, the urgency and whether a human is needed.
Support Ticket Triage (fan-out)	One call answers category, bug severity, repro steps, refund and frustration; AND rules decide which answers matter.
Voice Banking Commands (confidence-gated)	Classify a spoken banking command; risky actions need more confidence than read-only ones before acting.

💳 Pricing

Genesis is free — 1M decisions a month. Past the included volume, every plan keeps running on prepaid credits, and adding a card gives you $20 in free credits (valid 90 days). Plans and prices at dcision.io.

---

dcision.io · Docs · App · X · contact@dcision.io
_{Examples in this repository are MIT-licensed.}