Agno boundary

Type: agentic-system-analyses/types/agentic-system-boundary.md

Boundary and evidence

This run characterizes the agent execution, orchestration and retained-state mechanisms of Agno's Python SDK and AgentOS server runtime. It supports source-grounded comparison of how agents build context, invoke models and tools, coordinate other agents, and retain and reuse sessions, memory, learning and knowledge. It does not characterize the entire Agno product platform or every package in the monorepo.

The target is in scope because it owns model-dependent execution rather than merely returning a tool result to another runtime. Its target class is enclosing runtime: the local Agent run implementation prepares messages and tools and issues the model call. Team and Workflow provide additional orchestration entry points. This is a subsystem-only boundary within the repository, not a claim of whole-monorepo coverage. AgentOS is included, not treated as an external caller. The shipped cookbook drivers and their prompts are also included as consumers and configuration evidence. SRC-1: libs/agno/agno/agent/agent.py, libs/agno/agno/agent/_run.py, libs/agno/agno/team/team.py, libs/agno/agno/workflow/workflow.py, cookbook/00_quickstart/run.py.

The runtime's model-call responsibility is visible in the inspected run path:

model_response: ModelResponse = call_model_with_fallback( agent.model, agent.fallback_config, --- libs/agno/agno/agent/_run.py:554-556

The shipped AgentOS driver registers agents, a team and a workflow, obtains the app and serves it. AgentOS app construction includes database and optional scheduler and durable-queue lifespans. Its agent HTTP streaming adapter calls agent.arun and yields formatted run events. These are shipped runtime responsibilities inside this boundary, not responsibilities attributed to a hypothetical application. SRC-1: cookbook/00_quickstart/run.py, libs/agno/agno/os/app.py, libs/agno/agno/os/routers/agents/router.py.

app = agent_os.get_app() --- cookbook/00_quickstart/run.py

async for run_response_chunk in run_response: # type: ignore[union-attr] yield format_sse_event(run_response_chunk) # type: ignore --- libs/agno/agno/os/routers/agents/router.py

Retained-state production and consumption belong inside the functional boundary. Initial inspection traced session loading through database/cache selection, and persistence through separate session and run writes. The shipped storage example enables history context with a stable session ID; the memory example teaches preferences in one session and requests recall in another under the same user ID. The knowledge example inserts a local document and gives its Knowledge object to an agent with search enabled. Those examples establish shipped usage, not successful observed recall. SRC-1: libs/agno/agno/agent/_session.py, libs/agno/agno/agent/_run.py, cookbook/00_quickstart/agent_with_storage.py, cookbook/00_quickstart/agent_with_memory.py, cookbook/00_quickstart/agent_search_over_knowledge.py.

Persist the session row and this single run (both O(1))

_session.save_session(agent, session=session) _session.save_run( --- libs/agno/agno/agent/_run.py

The memory manager itself issues model calls with database tools. LearningMachine declares a unified interface for recall, processing, tool generation and maintenance; its detailed stores and curator remain available for later inspection. SRC-1: libs/agno/agno/memory/manager.py, libs/agno/agno/learn/machine.py.

Generate a response from the Model (includes running function calls)

response = model_copy.response( messages=messages_for_model, tools=_tools, ) --- libs/agno/agno/memory/manager.py:1095-1099

Functional scope and exclusions

The following map distinguishes inclusion from initial inspection. Available but uninspected included code is not an access gap or evidence for its behavior.

The separate package boundaries and console-entry-point distinction are grounded in SRC-1: libs/agno/pyproject.toml, libs/agnoctl/pyproject.toml, libs/agno_infra/pyproject.toml.

[project.scripts] agno = "agnoctl.main:app" --- libs/agno/pyproject.toml

External dependencies include model and embedding providers, database/vector-store services and their client libraries, tool-accessed services, FastAPI/Uvicorn and optional interface/telemetry integrations. The source includes their Agno-side adapters and configuration, not their independently implemented guarantees. SRC-1: libs/agno/pyproject.toml, cookbook/00_quickstart/agent_search_over_knowledge.py. The hosted AgentOS UI, online documentation and separately linked deployment-template repositories are outside the frozen source. The root introduction separates SDK, runtime and UI:

• Build your agent platform using the Agno SDK.
• Run your agent platform using the AgentOS runtime.
• Manage everything using the AgentOS UI. --- README.md

Those external exclusions prevent conclusions about browser-side operation, deployed platform configuration, current online documentation and end-to-end service behavior. No model calls, target execution or tests were performed for this boundary. Consequently it supplies implementation and declared-intent evidence, not observed operation, reliability measurements or causal experiments. There is no identified source-access gap for the selected subsystem; material not yet inspected remains explicitly uninspected.

The frozen evidence is repository https://github.com/agno-agi/agno at commit 00497422d4a386d93ab3b45c1a11faf42fcc932f. The acquired checkout's HEAD was verified against that pin. The applicability cutoff is the supplied run date, 2026-10-13; it does not imply that later changes or a running deployment match this checkout. Overall evidence tier: code-grounded.

Source register