Completed
Code Execution: The agent can generate and run code in Python, Node, Ruby, and Go on a one-off Dyno. It even supports installing dependencies on the fly. [27:02]
Class Central Classrooms beta
YouTube videos curated by Class Central.
Classroom Contents
Building Agentic Applications with Heroku Managed Inference and Agents
Automatically move to the next video in the Classroom when playback concludes
- 1 Introduction to Heroku AI [00:00]
- 2 Core Mission: The product's goal is to make every software engineer an AI engineer. Anush Dsouza, the Product Manager, states Heroku wants to make it “simple to attach agents and AI to your applicati…
- 3 Agentic Control Loop: Heroku provides an "agentic control loop" running on its platform. This loop gives AI models access to tools like code execution and data access, all secured under Heroku's trus…
- 4 AI Primitives: Heroku AI is built on key primitives. These include inference for accessing curated models, the Model Context Protocol MCP for extending app functionality, and PG Vector for handling e…
- 5 Trusted Compute: Heroku's trusted compute layer, Dynos, runs first-party tools. They plan to expand this with tools for web search and memory, and users can bring their own tools via MCP. [07:08]
- 6 Managed Inference: This service allows you to run AI models directly within your Heroku infrastructure. This keeps your data within your application's network for enhanced security. [13:23]
- 7 Supported Models: The platform supports text-to-text models from Anthropic Claude 3.5, 3.7, and 4, embeddings from Cohere Embed, and image generation with Stable Image Ultra. [14:38]
- 8 Chat Completions API: The basic chat completions endpoint is designed to be highly compatible with the OpenAI and Anthropic APIs. The presenter notes it's “95% compatible with the OpenAI API,” allowi…
- 9 Serverless Execution: Tools run on one-off Dynos, which scale to zero after execution. This means you “only pay for the compute that you use.” [17:57]
- 10 Dyno Run Command: This powerful tool allows the LLM to execute Unix commands or pre-deployed scripts on a Heroku Dyno. This gives the agent access to real-time information and the ability to interact…
- 11 postgres-get-schema: This retrieves the database schema, which helps prevent the LLM from hallucinating incorrect table or column names. [25:45]
- 12 postgres-run-query: This tool generates and executes SQL queries based on the provided schema and the user's natural language request. [25:52]
- 13 Code Execution: The agent can generate and run code in Python, Node, Ruby, and Go on a one-off Dyno. It even supports installing dependencies on the fly. [27:02]
- 14 Bring Your Own Tools: You can extend the agent's capabilities by deploying your own tools as MCPs to Heroku. [37:38]
- 15 Deployment: MCPs are deployed by configuring a Procfile with an mcp process type. This makes your custom tool discoverable by the Heroku agent. [46:19]
- 16 Example MCP: The workshop demonstrates a "Brave Search MCP" that allows the agent to perform web searches, showcasing how to add external knowledge to the agent. [43:42]
