Repository navigation
Binary Variants
Tier: Intermediate
qsv ships five binary variants. They share most code but differ in feature set, size, and intended use. Pick the right variant for your environment.
Tip
TL;DR: Use qsv (full) on your workstation. Use qsvlite if you're migrating from xsv or want a tiny install. Use qsvdp inside a DataPusher+ CKAN pipeline. Use qsvmcp to expose qsv as an MCP server. Use qsvpy* if you need the Python feature.
| Variant | Commands | Approx. size | Features | Use case |
|---|---|---|---|---|
qsv |
77–78 prebuilt (macOS / x86_64 Linux) | full | A per-platform feature list (see the note below), including get; a source build with -F all_features adds get_cloud, clipboard and more |
Day-to-day workstation use; the default |
qsvpy311/312/313/314 |
qsv + py
|
full + Python | Same as qsv plus the Python feature (dynamically linked) — adds py
|
When you need qsv py or Python expressions for transformations |
qsvmcp |
70 | smaller |
feature_capable, geocode, mcp, polars, profile, readstat, self_update, synthesize, to, viz
|
Running qsv as a Model Context Protocol server for Claude / other MCP clients |
qsvlite |
51 | ~16 % of qsv
|
None of the optional features | xsv migrants; minimal CI installs; low-memory environments |
qsvdp |
43 | ~16 % of qsv
|
DataPusher+ optimized: embedded luau, slim applydp, no progress bar, --update-gated self-update |
DataPusher+ CKAN pipelines |
Each figure includes the built-in help command. The counts come from running --list on the official prebuilts: qsv (77), qsvmcp (70) and qsvlite (51) on aarch64-apple-darwin; qsv (78), qsvmcp, qsvlite and qsvdp (43) on x86_64-unknown-linux-gnu. qsvdp is not shipped for macOS.
Note
The prebuilt qsv doesn't have every command. qsv has 80 commands across all its variants (the README table). The release workflows build qsv from an explicit, per-platform feature list that leaves out get_cloud and geoconnex; it also leaves out ui (clipboard) on macOS, Windows x86_64 and Linux x86_64/ARM64 (GNU and musl); only the Windows ARM64 (msvc and gnullvm), IBM Power (ppc64le) and IBM Z (s390x) builds include it. The aarch64-apple-darwin qsv lists 77 commands (no clipboard or prompt; applydp and py belong to qsvdp/qsvpy*); x86_64 Linux's feature list adds prompt (78, from the workflow, not a run). Other platforms vary — e.g. the Windows ARM64 builds use a smaller list that does include ui. Run qsv --list to see what your binary has. The prebuilt qsv includes get, so qsv get and dc: inputs work on every platform; cloud sources (s3://, gs://, az://) need qsvdp or a source build with -F get_cloud (Get & Disk Cache).
qsvmcp omits the 10 commands MCP doesn't need — apply, clipboard, color, fetch, fetchpost, foreach, get, lens, luau and prompt.
The feature_capable, qsvmcp, lite, and datapusher_plus features are mutually exclusive at compile time. You build one binary variant per build.
Note
qsvlite does not bundle the describegpt command. Use the full qsv variant if you need describegpt.
Note
viz is in qsv, qsvpy* (both as viz_static, which adds PNG/SVG/PDF/JPEG/WebP export) and qsvmcp (as plain viz — viz over MCP is HTML-only by design, so the browser-backed static-export path was never reachable there), but not in qsvlite or qsvdp. It is enabled across the little-endian publish targets only: viz is unsupported on big-endian (s390x, ppc64 BE) and is a hard compile_error! there rather than a silently broken binary. See Visualization.
Note
readstat and its inverse writestat are both in qsv, qsvpy* and qsvmcp, but not in qsvlite or qsvdp. It rides on polars, so like viz it is absent from the big-endian targets (s390x, ppc64 BE). See Conversion & I/O.
Warning
qsvmcp has no get command, accepts no dc: disk-cache-prefix inputs, and reads no cloud (s3://, gs://, az://) sources. qsvdp and source builds with the get_cloud feature have full cloud support (the prebuilt qsv has plain get but no cloud sources; see the note under the matrix). Within qsvmcp, the http(s)://, dathere:// and ckan:// schemes work, as does auto-decompression of compressed sources. See Get & Disk Cache and MCP Server.
The qsvpy builds carry qsv's feature set (including viz) plus python, except get.
Warning
The x86_64 prebuilts target x86-64-v3 and require AVX2/FMA/BMI2 (#4649) — qsv, qsvlite, qsvmcp, qsvpy*, the .deb and the MCPB bundle. qsvdp is the one x86_64 variant that stays portable, since DataPusher+ servers often run on VMs without AVX2. See Installation.
Getting Illegal instruction (SIGILL) on an old x86_64 CPU? On a CPU without AVX2 it is expected: use qsvdp, a 23.0.x release (the last without the AVX2 requirement), or build from source (cargo build --release --bin qsv -F all_features, which defaults to the portable baseline). ARM64 / IBM Power / s390x are unaffected.
There are no "portable" qsvp* binaries (qsvp, qsvplite, qsvpdp, qsvpmcp) in any release archive. If a script, installer or bookmark expects one, use the regular binary of the same variant (qsvp → qsv, qsvplite → qsvlite, qsvpdp → qsvdp, qsvpmcp → qsvmcp).
target-cpu=native is used only on targets built on matching hardware — Apple Silicon (.cargo/config.toml's [target.aarch64-apple-darwin]), Windows-on-ARM, ppc64le and s390x.
Note
ARM64 prebuilt targets: aarch64-unknown-linux-gnu, aarch64-unknown-linux-musl, aarch64-pc-windows-msvc and aarch64-pc-windows-gnullvm (#4721, #4722). The Linux ARM64 musl build uses the reduced feature set of the x86_64 musl build and the system allocator, and — like aarch64-unknown-linux-gnu — ships no qsvdp or qsvmcp. The gnullvm binaries link libunwind/libc++ statically, so they need only Windows system DLLs.
Are you migrating from xsv and want a drop-in replacement?
├── Yes → qsvlite
└── No
│
Are you embedding qsv inside DataPusher+ / CKAN?
├── Yes → qsvdp
└── No
│
Are you exposing qsv as an MCP server to an LLM client?
├── Yes → qsvmcp
└── No
│
Do you need `qsv py` (Python expressions)?
├── Yes → qsvpy311 / qsvpy312 / qsvpy313 / qsvpy314 (match your Python install)
└── No → qsv (the default)
qsvlite preserves xsv's command set and CLI ergonomics. Aliases:
mv qsvlite xsv # if you want the same binary name
xsv stats data.csvOr keep both:
xsv stats data.csv # original xsv behavior
qsvlite stats data.csv # same thing, more modern
qsv stats data.csv # full-featured: 48 stats vs xsv's 12See Comparison vs others for a feature-by-feature comparison and the xsv vs qsv stats wiki page for the stats column count breakdown.
For the exact set of features each variant ships, run:
qsv --versionThe output includes the version, the feature list, and the upstream library versions (Polars, jemalloc, etc.). See docs/PERFORMANCE.md#version-details for how to interpret the output.
For the canonical feature-flag list and how to compile each variant from source, see docs/FEATURES.md and the Installation → Compile from source section.
A few commands are variant-specific by design:
| Command | Available in | Why |
|---|---|---|
apply |
qsv, qsvpy*
|
Full NLP / date / currency suite (the standalone apply feature; not in qsvmcp, which ships feature_capable but not apply) |
applydp |
qsvdp only |
Slim version of apply for CKAN; trim, safen, cast |
lens, clipboard, prompt
|
UI feature group: qsv, qsvpy*
|
Interactive / desktop-only |
mcp |
qsvmcp only |
MCP server entry point |
py |
qsvpy* only |
Python is dynamically linked to a specific 3.x version |
pro |
qsv, qsvpy*
|
Bridges to the qsv pro API |
For the full mapping, see docs/FEATURES.md.
- Installation
-
docs/FEATURES.md— feature flag matrix (canonical) -
docs/INTERPRETERS.md— Python / Luau build details -
MCP Server —
qsvmcpdeployment - qsv pro Spotlight — graphical companion
- Integrations — DataPusher+, CKAN, DuckDB, Python notebooks
qsv — GitHub · Releases · Discussions · crates.io · qsv pro · Try it online · Benchmarks · DeepWiki · DOI · Licensed MIT · Third-party notices · AI Policy
qsv was made possible by datHere: standards-based, best-of-breed, open source solutions to make your Data Useful, Usable & Used.
Edit this page: Contributing to the Wiki
Home · Why qsv? · Tier legend
- Whirlwind Tour · Notebooks · Lessons & Exercises
- Benchmarks · Interactive Benchmarks
- Visualizations Gallery
- Neuro-symbolic, Localizable, Data Schematics
- Feature Flags · Goals/Non-goals · Testing
- Talks: NYC SOD 2022 · csv,conf,v8 · PyCon US 2025 · csv,conf,v9 · NYC SOD 2026
- "Have we achieved ACI?" series: 1 · 2 · 3 · 4
- Sponsor
- All Commands (index)
- Selection & Inspection
- Transform & Reshape
- Aggregation & Statistics
- Joins & Set Ops
- SQL & Polars
- Validation & Schema
- Metadata Profiling (profile)
- Conversion & I/O
- Geospatial
- Visualization (viz)
- HTTP & Web
- Get & Disk Cache
- Scripting (Luau / Python)
- Indexing, Compression & Diff
- AI & Documentation
- Recipes index
- Inspect an Unknown CSV
- Clean & Normalize
- Geographic Enrichment
- Date Enrichment
- CKAN Integration
- DCAT-US v3 for data.gov
- JSON Schema Validation
- Round-Trip SAS / Stata / SPSS
- Build a Data Pipeline
- Stats → Insights
- Data-Quality Signals with moarstats
- Group By
- Fetch & Cache
- Larger-than-RAM CSV
- Diff & Audit
- Multi-table Joins
- Synthesize Fake Data
- Performance Tuning
- Benchmarks
- Tuned csv Fork Benchmarks
- Security & Dependency Audits
- Environment Variables
- Stats Cache & Caching
- Lookup Tables
- MiniJinja Templating