Plans & Billing

Voicex uses a plan-based access control system. Each organization is linked to a plan that determines model access, agent limits, and feature availability.

---

Plan Tiers

	Free	Starter	Pro	Enterprise
Monthly Price	$0	$29	$99	$499
Annual Price	$0	$290	$990	$4,990
Max Agents	1	5	25	999
Concurrent Calls	2	10	50	500
Max Call Duration	5 min	30 min	60 min	120 min
Monthly Minutes	60	1,000	10,000	100,000
Custom Providers	No	1	5	100

---

Model Access by Plan

LLM Models

Model	Free	Starter	Pro	Enterprise
ollama/llama3.2:3b	Yes	Yes	Yes	Yes
groq/llama-3.3-70b-versatile	—	Yes	Yes	Yes
groq/llama-3.1-8b-instant	—	Yes	Yes	Yes
openai/gpt-4o-mini	—	—	Yes	Yes
openai/gpt-4o	—	—	Yes	Yes

TTS Models

Model	Free	Starter	Pro	Enterprise
edge/en-US-AriaNeural	Yes	Yes	Yes	Yes
edge/en-US-GuyNeural	Yes	Yes	Yes	Yes
elevenlabs/21m00Tcm4TlvDq8ikWAM (Rachel)	—	Yes	Yes	Yes
openai/alloy	—	—	Yes	Yes
openai/nova	—	—	Yes	Yes

STT Models

Model	Free	Starter	Pro	Enterprise
deepgram/nova-2	Yes	Yes	Yes	Yes

---

How Plan Access Works

Model Access Check

Every time an agent is created or updated with a global provider's model, the system checks:

planModelSet = plan.modelSets[category]   // e.g., plan.modelSets.llm
modelKey = `${providerKey}/${modelId}`    // e.g., "groq/llama-3.3-70b-versatile"
allowed = planModelSet.has(modelKey)       // O(1) Set lookup

Client provider models bypass this check entirely. If an org has their own OpenAI provider, they can use any model through it regardless of plan.

Plan Caching

Plans are cached in a 2-level cache for performance:

Request → L1 (in-memory Map, 1 min TTL)
  └── MISS → L2 (Redis, 10 min TTL)
              └── MISS → MongoDB

The cached plan includes pre-computed modelSets (JavaScript Set objects) for O(1) lookup:

interface CachedPlan extends Plan {
  modelSets: {
    llm: Set<string>;    // O(1) .has() check
    tts: Set<string>;
    stt: Set<string>;
  };
}

Cache invalidation: call invalidatePlanCache(planId) after updating a plan.

---

Agent Status & Plan Enforcement

When a user's plan changes (e.g., downgrade from Pro to Free), agents using models above the new plan are automatically paused — not deleted.

Status Computation

What Happens on Downgrade

1. User downgrades from pro to free
2. Agent using groq/llama-3.3-70b-versatile (requires starter+)
3. Next API call to GET /agents computes status: 'paused_plan'
4. Agent appears in playground but is disabled (non-selectable)
5. Agent detail page shows warning: "Model llama-3.3-70b-versatile requires Starter plan"
6. User can edit the agent to use ollama/llama3.2:3b (free tier) → agent becomes active again
7. Voice sessions check this at connection time and reject paused agents

---

Plan Features

The features field controls capability flags:

interface PlanFeatures {
  customProviders: boolean;      // Can client add their own providers?
  maxCustomProviders: number;    // How many client providers allowed?
}

Feature	Free	Starter	Pro	Enterprise
customProviders	false	true	true	true
maxCustomProviders	0	1	5	100

Frontend behavior:

• Free users see "Upgrade to use custom providers" on the providers page
• Starter users see the providers page but can only add 1 provider
• The "Add Provider" button is disabled when at limit

---

Custom Plans

Plans can be custom: true for specific clients (sales deals):

{
  slug: "acme-enterprise",
  name: "Acme Enterprise",
  custom: true,
  public: false,          // not shown on pricing page
  models: { /* ... */ },
  limits: { /* ... */ },
  features: { customProviders: true, maxCustomProviders: 50 }
}

Custom plans:

• Are created directly in MongoDB or via admin API
• Are not shown on the public pricing page (public: false)
• Can have any combination of models and limits
• Are linked to orgs the same way as standard plans

---

Public Plans API

For the landing page pricing section:

GET /api/plans

Returns all plans where public: true and custom: false, sorted by price. No authentication required.

---

Provider Costs

STT (Speech-to-Text)

Provider	Cost	Free Tier
Deepgram	$0.0058/min	$200 credit (~34K min)

LLM (Language Model)

Provider	Cost	Free Tier
Ollama	Free (self-hosted)	Unlimited
Groq	~$0.05-0.08/M tokens	Free tier available
OpenAI	~$0.15-0.60/M tokens	—

TTS (Text-to-Speech)

Provider	Cost	Free Tier	Production?
ElevenLabs	~$0.30/1K chars	10K chars/month	Yes
OpenAI	$15/1M chars	—	Yes
Edge TTS	Free	Unlimited	No (dev only)

Cost Estimate (1,000 minutes/month, Starter plan)

Service	Cost
Deepgram STT	$5.80
Groq LLM	~$0.50
ElevenLabs TTS	~$15-30
Total	~$21-36