How smartpipe compares

There are a lot of "LLM in your terminal" tools. This page maps where smartpipe fits, and where you might reach for something else.

Positioning

smartpipe brings semantic operators to Unix pipes: map, filter, reduce, cluster, join, and distinct.

It works over text and files: PDFs with figures, scans, images, audio, and video. It can use local Ollama models, or cloud providers you configure explicitly.

The landscape splits into two worlds, and smartpipe sits between the two:

• Semantic data frameworks (DocETL, LOTUS) have the exact operator set - map, filter, reduce, rank, group-by, automatic chunking. But you express pipelines in YAML or a pandas-style Python API and run them with a job runner. They batch-process files; they aren't stdin→stdout filters.
• Terminal LLM tools (llm, smartcat, mods, aichat, fabric, sgpt) are genuinely pipe-composable and often support local models. Almost all expose a single "prompt" verb rather than distinct operators.

smartpipe combines the operator model from the first world with the pipe ergonomics from the second.

At a glance

	Distinct verbs (map/filter/reduce/embed/rank)	stdin→stdout Unix filter	Local model path	Auto file parsing	Terminal-adaptive output
smartpipe	✅ all five	✅	Ollama autodetect; explicit cloud config	✅	✅
llm (Simon Willison)	partial (embed/similar)	✅	via plugin	-	-
smartcat	one verb	✅	✅	-	-
mods	one verb	✅	configurable	-	-
fabric	patterns, not operators	✅	✅	-	-
DocETL / LOTUS	✅ all	❌ (YAML/Python)	-	✅	❌

Where the other tools are genuinely better

Some tools fit certain jobs better:

• llm has a mature plugin ecosystem: many model backends, a logging database, and a large community. If you want backend breadth, llm is deeper.

• fabric ships hundreds of curated prompt "patterns." If you want a prompt library more than composable verbs, it is a better fit.

• DocETL / LOTUS have cost optimizers and a visual pipeline builder. For large declarative document jobs with a team, those frameworks are more powerful. smartpipe is for the command line, not a data platform.

• aichat is an all-in-one chat/RAG/session/web UI tool. smartpipe has no chat mode and no server.

What smartpipe does that's rare or unique

• Distinct semantic verbs as pipe stages. smartpipe has map, filter, reduce, embed, top_k, and more. It is not a single "prompt" command.

• Local model probing, with explicit cloud setup - if no chat model is configured, smartpipe probes local Ollama first. If it finds no usable model, it stops with setup instructions. Cloud calls happen only after you choose a cloud model or login path.

• grep --color=auto-style adaptive output - human view at a terminal, JSONL when piped, with no flag.

• Automatic recursive chunking in reduce - summarize an input far larger than the model's context, with no configuration.

• Automatic file parsing - point smartpipe at PDFs; you never name a parser.

• No tool-use surface - smartpipe doesn't execute model output; a response is treated as data, not commands (see privacy).

When not to use smartpipe

• You want a chat REPL → aichat, llm chat.
• You want a big library of prewritten prompts → fabric.
• You're building a large declarative document pipeline with joins and a UI → DocETL.
• You need a model backend smartpipe doesn't support and can't reach via the OpenAI-compatible endpoint → llm with the right plugin.

The tool landscape moves monthly; this page reflects the survey behind smartpipe's design at the time of writing. Corrections welcome.