Lattice
Skip to content

11 comparisons

When the list is not the answer.

A directory can tell you what exists. It cannot tell you which of two things to pick. These comparisons cover tools that are genuine substitutes for one another, and end with a recommendation rather than a feature grid.

Not seeing your pair? Build your own comparison from any three tools, in any layers.

Across layers · 3

Tools in different layers that compete for the same week of work. No vendor publishes these, because no vendor sells every column. Each ends in an order of adoption, not a winner.

  1. Gateway vs guardrails vs evals: what to build first

    A gateway, a guardrail layer and an eval suite compete for the same first week of work. What each catches, what each misses, and the order to adopt them in.

    LiteLLM · Guardrails AI · promptfoo

  2. Retrieval vs fine-tuning vs prompt optimisation: fixing wrong answers

    Three layers that each claim to fix wrong answers, compared on what they actually change, what they need from you, and which failure each one is for.

    pgvector · Unsloth · DSPy

  3. Self-hosting vs routing vs measuring: cutting inference cost

    Three ways to lower an inference bill — run the model yourself, send easy queries to a smaller model, or measure where spend goes — compared on savings and risk.

    vLLM · RouteLLM · Helicone

Substitutes · 8

Tools that do the same job. Pick one.

  1. vLLM vs SGLang vs TGI vs llama.cpp

    The four open inference runtimes compared on prefix reuse, hardware floor, structured output and operational complexity — with a recommendation for each traffic shape.

    vLLM · SGLang · Text Generation Inference · llama.cpp

  2. Langfuse vs LangSmith vs Braintrust vs Phoenix

    Four LLM observability and evaluation platforms compared on self-hosting, eval primitives, open standards and how much they lock you in.

    Langfuse · LangSmith · Braintrust · Arize Phoenix

  3. pgvector vs Qdrant vs Pinecone vs Chroma

    Vector stores compared on hybrid search, filtering, operational burden and when the honest answer is that you should not add one at all.

    pgvector · Qdrant · Pinecone · Chroma

  4. Temporal vs Inngest vs Trigger.dev vs Restate

    Durable execution engines compared on language, replay model and whether they fit inside your existing deployment — the layer that decides whether an agent survives a deploy.

    Temporal · Inngest · Trigger.dev · Restate

  5. LiteLLM vs Portkey vs Cloudflare AI Gateway

    LLM gateways compared on provider coverage, caching, where the data goes and whether the routing logic is yours or theirs.

    LiteLLM · Portkey · Cloudflare AI Gateway · Envoy AI Gateway

  6. Instructor vs JSON mode vs constrained decoding

    Three ways to get a schema-valid response from a model, compared on reliability, portability and whether the guarantee is real.

    Instructor · Guidance · Outlines

  7. LangChain vs LlamaIndex vs Pydantic AI vs OpenAI Agents SDK

    Agent frameworks compared on retrieval depth, type safety, surface area and ecosystem — with a recommendation for data-heavy, schema-heavy and minimal-loop teams.

    LangChain · LlamaIndex · Pydantic AI · OpenAI Agents SDK

  8. NeMo Guardrails vs Guardrails AI vs Presidio vs Invariant

    Safety layers compared on input rails, output validation, PII handling and whether they run inline or as a separate service — with a recommendation by compliance shape.

    NeMo Guardrails · Guardrails AI · Microsoft Presidio · Invariant Guardrails