Agents in your app

Agents are assistants you build into your own product. They run on your own model provider account (OpenAI, Anthropic, Google or any OpenAI-compatible API) and act with the permissions of the person using them, so row-level security decides what each one sees.

#Build an agent

1. 1

   Open Agents

   Describe what you want and Sovariq drafts the name, instructions and tools, or start from a template or blank. You review everything before saving.

2. 2

   Model and key

   Pick a provider, paste your API key (encrypted, never shown again) and choose a model. You pay the provider directly.

3. 3

   Data access

   No direct SQL, Read any table (list, describe, SELECT) or Read and write (also INSERT/UPDATE/DELETE, asking before each change). Queries run as the signed-in person.

4. 4

   Knowledge

   Pick knowledge bases (Knowledge bases (RAG)): the agent searches them and answers with numbered citations like [1], or says it doesn't know. It searches as the person using it, so each user only gets passages they may read.

5. 5

   Tools and playbooks

   Add tools: one parameterised SELECT, one INSERT/UPDATE/DELETE (with a WHERE), a call to a public HTTPS API, or one of your edge functions. Playbooks (up to 5) are step-by-step rules for specific situations.

6. 6

   Who can use it

   Your team, or your app's signed-in users. Set Runs per user per day for app users.

#Run it from your app

Call the agent with the signed-in user's session; the SDK streams the reply as it is written. Pass the threadId from a reply to continue that conversation: each user only sees their own threads, and every turn uses the agent's current instructions and limits.

const reply = await sov.agents.run(agentId, {
  message: 'Where is my order 1042?',
  onEvent: (e) => { if (e.type === 'text_delta') render(e.text) },
})
await sov.agents.run(agentId, { message: 'And the invoice?', threadId: reply.threadId })
const threads = await sov.agents.threads(agentId)   // this user's conversations

#Limits

Runs per user per day (default 50) caps how many runs each signed-in person can start each day. Past the cap the endpoint answers 429 until the next day (UTC). This stops one user from spending your model account and your plan's monthly agent runs for everyone else. Set it to 0 for no per-user cap; the monthly limit still applies. Agents for app users never get tools that delete data (Sovariq safety, approvals and guard rails).

#Coming from Firebase or Supabase?

Moving an existing app? Migrate from Supabase or Firebase imports users (with their passwords), data, files and rules.