Agentivium Core

Shared infrastructure for research systems.

Agentivium Core captures reusable contracts for intent, evidence, policy, state, execution, and evaluation without embedding domain-specific logic. It supplies schemas and interfaces for research systems, rather than a complete agent runtime or orchestration platform.

Different research systems repeatedly need the same basic mechanisms. Core makes these contracts reusable; domain extensions implement their own policies, tools, and execution behavior.

Python package ↗

Intent, policy, and planning

Typed intent schemas and interfaces for parsing requests, validating policy, and building action plans.

Tools and LLM calls

Tool adapter contracts and provider-neutral request, response, and structured-output interfaces.

State, memory, and retrieval

Conversation state, context packs, scoped memory records, and retrieval interfaces for multi-turn assistants.

Verification

Contracts for evidence, verification checks, intent-decision relations, and admissibility before domain execution.

Tracing and evaluation

Execution traces, LLM call traces, and evaluation records for inspecting and comparing system behavior.