Exploiting the plasticity of human brain

Kaustubh kausmeows

Exploiting the plasticity of human brain

Engineering @agno-agi | Research, University of Auckland, NZ

kausmeows / runs-fallback.json

Created March 23, 2026 11:33

	[
	{
	"input": {
	"input_content": "Call the break_model tool"
	},
	"model": "gpt-4o",
	"tools": [
	{
	"result": "Tool executed successfully.",
	"metrics": {

kausmeows / team-hitl-db.json

Created March 18, 2026 12:45

	[
	{
	"input": {
	"input_content": "Deploy the payments app version 2.1 to production. Context: The user wants v2.1 deployed to the production environment. Assume you have access to the CI/CD pipelines, production cluster, and necessary credentials. Steps to perform and expected outputs:\n\n1) Confirm the code artifact for payments app v2.1 is available (build artifact or container image). Provide artifact SHA/tag and location (registry/repo).\n2) Run pre-deployment checks: verify health of production cluster, ensure there are no ongoing deployments or blocking incidents, check required service/DB migrations and schedule if needed.\n3) Trigger the production deployment using the established pipeline. If the pipeline requires manual approval, report that and provide instructions or request approval from the user.\n4) Apply any required database migrations as part of the deployment; run in a safe manner (e.g., migrate with rollback plan). If migrations are destructive, pause and ask for explicit

kausmeows / parser_native_structured_output.py

Created March 11, 2026 13:29

	"""
	Parser Model — Structured Output Debug
	=======================================

	Proves that:
	- Primary model call: response_format is None (no structured output params)
	- Parser model call: response_format contains the full JSON schema (native structured output)
	"""

	import json

kausmeows / custom_retriever_with_run_context.py

Created March 10, 2026 13:41

	"""
	Custom Retriever with RunContext
	=============================

	Demonstrates how to pass application-controlled data (like a project_id
	or file_name) into a custom retriever using RunContext.dependencies.

	This is useful when:
	- Your retriever needs scoping (e.g., per-project, per-tenant)
	- You want to pass runtime filters that the LLM shouldn't control

kausmeows / hitl-continue-api-test.py

Created March 3, 2026 09:25

	"""
	Test Team HITL Continue Run API
	===============================

	This script tests the team continue run endpoint by:
	1. Starting an AgentOS server with a team that has a tool requiring confirmation
	2. Creating a run via the API (which will pause)
	3. Continuing the run via the /continue API endpoint
	"""

kausmeows / custom_event_team.py

Created February 27, 2026 05:11

	"""
	Example: CustomEvents with Teams (respond_directly=False)

	This demonstrates how CustomEvents from member agents are streamed
	to the user-facing stream even when respond_directly=False.
	"""

	import time
	from agno.agent import Agent
	from agno.team import Team

kausmeows / steps_function.py

Created February 2, 2026 07:11

	from typing import AsyncIterator, Union

	from agno.agent import Agent
	from agno.db.in_memory import InMemoryDb
	from agno.models.anthropic import Claude
	from agno.os import AgentOS
	from agno.run.workflow import WorkflowRunOutputEvent
	from agno.workflow import Step, Workflow
	from agno.workflow.steps import Steps
	from agno.workflow.types import StepInput, StepOutput

kausmeows / mcp_errors.py

Created February 1, 2026 18:32

MCP tool failure isolation in AgentOS

	"""
	Example demonstrating MCP tool failure isolation in AgentOS.

	This example shows that when an AgentOS app includes multiple agents - some with MCP tools
	and some without - a connection failure to an MCP server should NOT cause the entire app
	to fail. Only the agent using the failing MCP tool should be affected.

	Run this example with:
	python cookbook/05_agent_os/mcp_demo/mcp_tools_failure_isolation.py

kausmeows / workflow_with_bg_hook.py

Created January 9, 2026 05:01

	"""
	Example: Workflow with Conversational Intake and Background Research

	This demonstrates a Workflow where:
	1. An intake agent step has a conversation with the user
	2. Once the user's name is captured, background research kicks off
	3. The research results are stored in the workflow's shared session_state
	4. Subsequent workflow steps/turns can access the research

	Run with: python libs/agno/agno/test.py

kausmeows / workflow_input_schema.py

Created January 2, 2026 08:59

	# paper2saas_agent.py
	import argparse
	import os
	import json
	from typing import Dict, Any, List

	from pydantic import BaseModel, Field

	# Agno SDK Imports
	from agno.workflow import Workflow, Step, Parallel