Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.samvad.dev/llms.txt

Use this file to discover all available pages before exploring further.

The protocol defines one authentication hook and gets out of your way. Billing, key issuance, and subscription logic live in your product — not in SAMVAD.

​
Public service

Open to anyone with a valid signature. The agent owner absorbs all LLM costs. Use tokensPerSenderPerDay to prevent any single caller from exhausting your budget. 
agent.skill('summarize', {
  trust: 'public',
  // ...
})

// In agent card:
"rateLimit": {
  "requestsPerMinute": 60,
  "requestsPerSender": 10,
  "tokensPerSenderPerDay": 50000
}
When a sender exhausts their daily budget, the agent returns TOKEN_BUDGET_EXCEEDED with a Retry-After header pointing to UTC midnight reset.

​
Internal / multi-agent orchestration

Restrict access to a specific set of internal agent identities. Only the listed agent:// IDs can call the skill, verified by signature. 
agent.skill('process-order', {
  trust: 'trusted-peers',
  allowedPeers: [
    'agent://billing.internal',
    'agent://inventory.internal',
  ],
  // ...
})
Use AgentClient.prepare() on each internal agent to get its agentId and publicKey, then register them with agent.trustPeer() before starting the server.

​
Commercial / paywalled

Require a Bearer token for access. You issue tokens from your own paywall or subscription flow — the protocol just enforces their presence. The issuance, validation, and revocation logic is entirely yours. 
agent.skill('generate-report', {
  trust: 'authenticated',
  // ...
})
Callers include the token in every request: 
{
  "auth": { "scheme": "bearer", "token": "your-issued-token" },
  "payload": { "…": "…" }
}
The SDK checks that the auth.token field is non-empty for authenticated skills. What it contains and how you validate it in your handler is up to you.

​
Mixing tiers

An agent can expose skills at different tiers — public diagnostics, authenticated core functionality, and trusted-peer internal endpoints: 
agent.skill('health-check', { trust: 'public', ... })
agent.skill('analyze', { trust: 'authenticated', ... })
agent.skill('admin-reset', { trust: 'trusted-peers', allowedPeers: ['agent://ops.internal'], ... })

​
What the protocol doesn’t do

  • Issue, rotate, or revoke API keys
  • Implement subscriptions or usage-based billing
  • Validate what’s inside a Bearer token (that’s your handler’s job)
  • Manage payment between agents
The protocol provides one primitive per tier. Everything built on top of those primitives is your product.