Aezab

Agent infrastructure

Self-hosted agent infrastructure for orchestration, hybrid retrieval, workflows, tools, and full-trace audit behind one API.

Public repository

Purpose

Aezab combines a backend service, web console, and headless API. Teams configure and test agents in the console, then embed them into existing applications, support systems, internal tools, or automation pipelines through the API.

Agents, knowledge sources, workflows, and tools are stored as runtime configuration rather than compiled into application code.

Runtime

Knowledge search, workflows, HTTP tools, and agent-to-agent delegation are exposed as function-calling tools. The model chooses among them directly, with supporting heuristics alongside those calls rather than a router deciding first.

Capability unit: Skills: workflow, tool_call, knowledge_qa, delegate, and composite.
Deployment: Self-hosted and multi-tenant, with model configuration resolved from agent binding to tenant and environment defaults.

Permissions and execution

Tools are registered with typed input and output contracts, authentication, execution policy, a required permission scope, and a declared risk level. Workflows provide typed steps for procedures that should not be improvised.

Tool rounds: MAX_TOOL_ROUNDS = 5
Delegation: MAX_DELEGATION_DEPTH = 3
External services: A circuit breaker is applied per service.
Record: Each invocation carries a trace_id.