A agentindex
Home · stack

Build my own agent

I want to put an AI agent into my app. Here's a stack that does this job, and a worked example of it in use.

Case study — Solo builder

A solo builder ships a product with an AI feature

A solo builder is shipping a product and wants to put an AI agent into it. They have no team and no time to rebuild what others already solved, so they assemble a stack rather than write one.

Their stack has four pieces. Their application holds the business logic. An agent framework — a library — wires that logic, the tools, and the prompts into an agent, so they don’t hand-roll orchestration. A model provider supplies the brain, and because cost matters at this stage, they bring their own key to control it. Once the agent is live, tracing lets them see what it actually did when something breaks.

Field accounts of agentic coding workflows agree on the shape: pick composable pieces, keep the model swappable, and add observability before you need it rather than after.

Sources

  • Addy Osmani on LLM workflows (raw/0012-addyosmani-llm-workflow.md)
  • State of agentic coding, session 1 (raw/0017-armin-state-of-agentic-coding-1.md)

The stack

Click a box to filter the list to just the tools that fill it.

flowchart LR
  app(("Your application"))
  framework["Agent framework"]
  model["The model"]
  observe["See what it does"]
  app -->|calls| framework
  framework -->|sends prompts to| model
  framework -->|emits traces to| observe

Agent framework

The library that wires your code, tools, and prompts into an agent.

  • Agno — A high-performance multi-agent framework and runtime (formerly Phidata).
  • AutoGen — Microsoft's framework for multi-agent conversations.
  • AWS Strands Agents — AWS's model-driven, open-source agent SDK.
  • Cloudflare Agents SDK — Stateful agents on Cloudflare Workers.
  • CrewAI — Role-playing multi-agent crews.
  • DSPy — Declarative, programmable prompt programming from Stanford.
  • Google Agent Development Kit — Google's code-first toolkit for production agents.
  • Haystack (deepset) — Production-grade RAG and agent pipelines from deepset.
  • LangChain — The broad framework for LLM applications and agents.
  • LangGraph — Graph-based agent orchestration from LangChain.
  • Letta — Stateful agents with long-term memory (formerly MemGPT).
  • LlamaIndex — The data framework for LLM applications.
  • Mastra — The TypeScript-first agent framework.
  • OpenAI Agents SDK — OpenAI's lightweight agents SDK (Swarm successor).
  • PocketFlow — A 100-line minimalist LLM framework, agentic-coding friendly.
  • Pydantic AI — Type-safe agent framework from the Pydantic team.
  • Semantic Kernel — Microsoft's SDK for enterprise AI orchestration.
  • Smolagents — Hugging Face's minimalist code-writing agents.
  • Vercel AI SDK — The TypeScript toolkit for building AI apps and agents.
See all on the directory →

The model

The brain the agent calls. Bring your own key or pick a provider.

  • Anthropic — The Claude model API.
  • Cerebras Inference — Wafer-scale inference for very high throughput.
  • Cohere — Enterprise LLMs, embeddings, and rerankers.
  • DeepInfra — Low-cost hosted inference for open models.
  • Fireworks AI — Fast hosted open-model inference.
  • Google AI Studio — The Gemini API with a generous free tier.
  • Groq — Ultra-low-latency inference on custom LPU silicon.
  • llama.cpp — C/C++ inference for GGUF models on any hardware.
  • LMDeploy — Efficient inference and serving for open models.
  • Mistral AI — Provider of the Mistral and Codestral models.
  • Novita AI — Affordable hosted LLM and image APIs.
  • NVIDIA NIM — NVIDIA's inference microservices for hosting open and frontier models.
  • Ollama — Run large language models locally with one command.
  • OpenAI — The GPT and Codex model API.
  • OpenRouter — One API routing to hundreds of models across providers.
  • Replicate — Run and host thousands of open models via API.
  • SGLang — Fast serving and structured generation engine.
  • Text Generation Inference — Hugging Face's production LLM server.
  • Together AI — Hosted open-model inference and fine-tuning.
  • vLLM — High-throughput self-hosted LLM serving engine.
  • Z.ai — Provider of the GLM family of models.
See all on the directory →

See what it does

Tracing so you can watch what the agent actually did in production.

  • AgentOps — Observability built specifically for AI agents.
  • Arize Phoenix — Open-source LLM tracing and eval.
  • Comet Opik — Open-source eval and tracing from Comet.
  • Fiddler — AI observability and governance for enterprise.
  • Helicone — Proxy-based LLM observability with instant cost tracking.
  • Laminar — Open-source observability and eval for AI agents.
  • Langfuse — The most-used open-source LLM observability tool.
  • LangSmith — LangChain's observability and eval platform.
  • Lunary — Open-source LLM observability and prompt management.
  • Pydantic Logfire — Observability from the Pydantic team.
  • SigNoz — Open-source OpenTelemetry observability.
  • Weights and Biases Weave — LLM tracing and eval inside W&B.
See all on the directory →