Sovariq AI and agents in your app

Sovariq is the assistant in the console: describe what you want ("a table of orders only their owner can see") and it plans the change, shows it, and runs it with your permissions after you approve. It uses the same public API as the SDKs and the CLI — there is no back door — so policies, plan limits and the audit trail apply.

Agents are assistants you build into your own product — a support bot for your customers, an analyst for your team. They run on your own model provider account and act with the permissions of the person using them.

#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 (OpenAI, Anthropic, Google or any OpenAI-compatible API), 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, so row-level security decides what the agent sees.

4. 4

   Tools and playbooks

   Add tools — one parameterised SELECT, one INSERT/UPDATE/DELETE (with a WHERE clause), 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.

#Run an agent from your app

Set Who can use it to your app's users, then call the run endpoint with the signed-in user's access token. The reply streams as server-sent events.

curl -N -X POST https://<project-ref>.sovaria.in/api/v1/sovariq/agents/$AGENT_ID/runs \
  -H "Authorization: Bearer $USER_ACCESS_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"message": "Where is my order 1042?"}'

#Safety

• Tool values are always bound as literals — the model can never change the SQL.
• API tools have a fixed host; private networks, IP literals and redirects are refused. Auth header values are encrypted.
• Write tools ask for approval unless you allow them to run on their own; destructive Sovaria tools always ask.

#Use Sovaria from your own AI assistant

sov mcp runs an MCP server so Claude, Cursor and other MCP clients can use the same Sovaria tools with your login — see CLI: Logs, vectors and AI tools.

#Plans