AI integration

Aksara's stable AI-native integration is generated MCP execution. Provider calls, natural-language query planning, and Studio AI remain experimental. There is no public QueryEngine, InsightGenerator, Planner, or AgentRuntime class.

Stable path: expose an authorized tool

Define an AI-exposed model and ModelViewSet, configure server-side Principal resolution, and connect an official MCP client to /mcp/. The framework generates schemas and executes calls through the same permission, policy, field, tenant, transaction, and ORM path as REST.

Follow the complete MCP quickstart.

Experimental structured query plans

An external model may produce an AiQueryPlan; Aksara can validate and execute that structure:

from aksara.ai.query import AiFilterCondition, AiQueryPlan, execute_ai_query_plan

plan = AiQueryPlan(
    model="Task",
    filters=[AiFilterCondition(field="done", lookup="exact", value=False)],
)
result = await execute_ai_query_plan(plan)

This low-level function does not resolve a request Principal. Keep it behind application authorization and tenant context. It does not translate natural language itself.

Experimental prompt providers

Configure the current AI Hub path for optional provider-backed Studio or prompt-pack features:

aksara ai-hub configure openai
aksara ai-hub status
aksara ai-hub doctor

from aksara.ai.runtime import run_prompt_pack

result = await run_prompt_pack(
    {
        "system_prompt": "Answer concisely.",
        "user_prompt": "Summarize the supplied application context.",
        "provider": "ollama",
        "model": "llama3",
    }
)

Treat provider output as untrusted. Provider selection and quality, session persistence, autonomous workflows, memory, and durable orchestration are not stable v0.7 guarantees.