Liking cljdoc? Tell your friends :D

CircleCI Clojars Project Financial Contributors on Open Collective project chat

Small Clojure Interpreter


Use from Clojure(Script)

(require '[sci.core :as sci])
(sci/eval-string "(inc 1)") => ;; 2
(sci/eval-string "(inc x)" {:namespaces {'user {'x 2}}}) ;;=> 3

Try SCI in your browser at NextJournal.

For usage with GraalVM native-image check here.


You want to evaluate code from user input, or use Clojure for a DSL inside your project, but eval isn't safe or simply doesn't work.

This library works with:

  • Clojure on the JVM
  • Clojure compiled with GraalVM native
  • ClojureScript, even when compiled with :advanced, and JavaScript

Projects using sci

Sci is used in:

  • 4ever-clojure. 4clojure as a static web page.
  • Babashka. A Clojure scripting tool that plays well with Bash.
  • Bootleg. An HTML templating CLI.
  • Bytefield-svg. NodeJS library to generate byte field diagrams.
  • Cardigan Bay. Wiki engine in Clojure.
  • Chlorine. Socket-REPL and nREPL package for Atom editor.
  • Clj-kondo. A Clojure linter that sparks joy.
  • Closh. Bash-like shell based on Clojure. GraalVM port is work in progress.
  • Dad. A configuration management tool.
  • Datalevin. Durable Datalog database.
  • Firn. Org-mode static site generator.
  • For-science. Discord bot.
  • Jet. CLI to convert between JSON, EDN and Transit.
  • Keycloak-clojure. Clojure library for Keycloak.
  • Logseq. A local-only outliner notebook which supports both Markdown and Org mode.
  • Malli. Plain data Schemas for Clojure/Script.
  • PCP. Clojure Processor (PHP replacement).
  • PGMig. Fast Standalone PostgreSQL Migration Runner.
  • Prose. Alternate syntax for Clojure, similar to what Pollen brings to Racket.
  • SICMUtils. Computer Algebra System in Clojure, tailored for math and physics investigations.
  • Spire. Pragmatic provisioning using Clojure.
  • Zprint. Tool to beautifully format Clojure(script) code and data.

Are you using SCI in your company or projects? Let us know here.


Experimental. Breaking changes are expected to happen at this phase. They will be documented in the


Use as a dependency:

Clojars Project

API docs

See the generated codox documentation.


The main API function is sci.core/eval-string which takes a string to evaluate and an optional options map.

In sci, defn does not mutate the outside world, only the evaluation context inside a call to sci/eval-string.

By default SCI only enables access to most of the Clojure core functions. More functions can be enabled by using :namespaces and :classes. Normally you would use SCI's version of println but here, for the purposes of demonstration, we use use Clojure's version of println instead:

user=> (require '[sci.core :as sci])
user=> (sci/eval-string "(println \"hello\")" {:namespaces {'clojure.core {'println println}}})

It is also possible to provide namespaces which can be required inside a SCI program:

user=> (def opts {:namespaces {' {'println println}}})
user=> (sci/eval-string "(require '[ :as lib]) (lib/println \"hello\")" opts)

You can provide a list of allowed symbols. Using other symbols causes an exception:

user=> (sci/eval-string "(inc 1)" {:allow '[inc]})
user=> (sci/eval-string "(dec 1)" {:allow '[inc]})
ExceptionInfo dec is not allowed! [at line 1, column 2]  clojure.core/ex-info (core.clj:4739)

Providing a list of disallowed symbols has the opposite effect:

user=> (sci/eval-string "(inc 1)" {:deny '[inc]})
ExceptionInfo inc is not allowed! [at line 1, column 2]  clojure.core/ex-info (core.clj:4739)


Providing a macro as a binding can be done by providing a normal function that:

  • has :sci/macro on the metadata set to true
  • has two extra arguments at the start for &form and &env:
user=> (def do-twice ^:sci/macro (fn [_&form _&env x] (list 'do x x)))
user=> (sci/eval-string "(do-twice (f))" {:bindings {'do-twice do-twice 'f #(println "hello")}})

Alternatively you can refer to the macro from the Clojure environment via the var (this only works in a JVM environment):

user=> (defmacro do-twice [x] (list 'do x x))
user=> (sci/eval-string "(do-twice (f))" {:namespaces {'user {'do-twice #'do-twice 'f #(println "hello")}}})


To remain safe and sandboxed, SCI programs do not have access to Clojure vars, unless you explicitly provide that access. SCI has its own var type, distinguished from Clojure vars.

In a SCI program these vars are created with def and defn just like in normal Clojure:

(def x 1)
(defn foo [] x)
(foo) ;;=> 1
(def x 2)
(foo) ;;=> 2

Dynamic vars with thread-local bindings are also supported:

(def ^:dynamic *x* 1)
(binding [*x* 10] x) ;;=> 10
(binding [*x* 10] (set! x 12) x) ;;=> 12
x ;;=> 1

Creating SCI vars from Clojure can be done using sci/new-var:

(def x (sci/new-var 'x 10))
(sci/eval-string "(inc x)" {:namespaces {'user {'x x}}}) ;;=> 11

To create a dynamic SCI var you can set metadata or use sci/new-dynamic-var:

(def x1 (sci/new-var 'x 10 {:dynamic true}))
(sci/eval-string "(binding [*x* 12] (inc *x*))" {:namespaces {'user {'*x* x1}}}) ;;=> 13
(def x2 (sci/new-dynamic-var 'x 10))
(sci/eval-string "(binding [*x* 12] (inc *x*))" {:namespaces {'user {'*x* x2}}}) ;;=> 13

SCI vars can be bound from Clojure using sci/binding:

(def x (sci/new-dynamic-var 'x 10))
(sci/binding [x 11] (sci/eval-string "(inc *x*)" {:namespaces {'user {'*x* x2}}})) ;;=> 11

The dynamic vars *in*, *out*, *err* SCI a sci program correspond to the dynamic sci vars sci/in, sci/out and sci/err in the API. These vars can be rebound as well:

(def sw (
(sci/binding [sci/out sw] (sci/eval-string "(println \"hello\")")) ;;=> nil
(str sw) ;;=> "hello\n"

A shorthand for rebinding sci/out is sci/with-out-str:

(sci/with-out-str (sci/eval-string "(println \"hello\")")) ;;=> "hello\n"

Copy a namespace

To copy the public vars of a Clojure namespace and to reify the Clojure vars into corresponding SCI vars, you can use ns-publics in Clojure and the following API functions:

  • sci/create-ns: creates an object that identifies a SCI namespace and carries the metadata of a SCI namespaces
  • sci/copy-var: macro that copies a Clojure var to a SCI namespace (created through sci/create-ns). Automatically converts dynamic vars and macros. Captures docstrings and arglists.

E.g. given the following Clojure namespace:

(ns foobar)

(defmacro do-twice [x] (list 'do x x))

(defn times-two [x]
  (* x 2))

you can re-create that namespace in a SCI context like this:

(require 'foobar)

(def fns (sci/create-ns 'foobar-ns nil))

(def foobar-ns {'do-twice (sci/copy-var foobar/do-twice fns)
                'times-two (sci/copy-var foobar/times-two fns)})

(def ctx (sci/init {:namespaces {'foobar foobar-ns}}))

(sci/binding [sci/out *out*]
  (sci/eval-string* ctx "(foobar/do-twice (prn :x))"))

(sci/eval-string* ctx "(foobar/times-two 2)")

To copy an entire namespace without enumerating all vars explicitly with sci/copy-var you can use the following approach using ns-publics and sci/new-var:

(reduce (fn [ns-map [var-name var]]
            (let [m (meta var)
                  no-doc (:no-doc m)
                  doc (:doc m)
                  arglists (:arglists m)]
              (if no-doc ns-map
                  (assoc ns-map var-name
                         (sci/new-var (symbol var-name) @var
                                      (cond-> {:ns fns
                                               :name (:name m)}
                                        (:macro m) (assoc :macro true)
                                        doc (assoc :doc doc)
                                        arglists (assoc :arglists arglists)))))))
          (ns-publics 'foobar))

Stdout and stdin


To enable printing to stdout and reading from stdin you can SCI-bind sci/out and sci/in to *out* and *in* respectively:

(sci/binding [sci/out *out*
              sci/in *in*]
  (sci/eval-string "(print \"Type your name!\n> \")")
  (sci/eval-string "(flush)")
  (let [name (sci/eval-string "(read-line)")]
    (sci/eval-string "(printf \"Hello %s!\" name)
                     {:bindings {'name name}})))
Type your name!
> Michiel
Hello Michiel!

When adding a Clojure function to SCI that interacts with *out* (or *in* or *err*), you can hook it up to SCI's context. For example, a Clojure function that writes to *out* can be Clojure bound to SCI's out:

user=> (defn foo [] (println "yello!"))
user=> ;; without binding *out* to sci's out, the Clojure function will use its default *out*:
user=> (sci/eval-string "(with-out-str (foo))" {:namespaces {'user {'foo foo}}})
;; Let's hook foo up to SCI's context:
user=> (defn wrapped-foo [] (binding [*out* @sci/out] (foo)))
user=> (sci/eval-string "(with-out-str (foo))" {:bindings {'foo wrapped-foo}})

To always enable printing in your SCI environment you can set sci/out to *out* globally:

(sci/alter-var-root sci/out (constantly *out*))


Similar to Clojure vs. CLJS, the difference wit SCI on Clojure vs. SCI on CLJS is that in the latter you should use sci/print-newline and sci/print-fn to control printing to stdout:

cljs.user=> (def output (atom ""))
cljs.user=> (sci/binding [sci/print-newline true sci/print-fn (fn [s] (swap! output str s))] (sci/eval-string "(print :hello) (println :bye)"))
cljs.user=> @output

This is supported since SCI 0.2.7 (currently unreleased but available as git dep).

To always enable printing in your SCI environment you can set sci/print-fn to *print-fn* globally:

(sci/alter-var-root sci/print-fn (constantly *print-fn*))


Creating threads with future and pmap is disabled by default, but can be enabled by requiring sci.addons.future and applying the sci.addons.future/install function to the sci options:

   [sci.core :as sci]
   [sci.addons.future :as future]))

(sci/eval-string "@(future (inc x))"
                 (-> {:namespaces {'user {'x 1}}}
;;=> 2

For conveying thread-local SCI bindings to an external future use sci/future:

   [sci.core :as sci]
   [sci.addons.future :as future]))

(def x (sci/new-dynamic-var 'x 10))

@(sci/binding [x 11]
     (sci/eval-string "@(future (inc x))"
                      (-> {:namespaces {'user {'x x}}}
;;=> 12


Adding support for classes is done via the :classes option:

(sci/eval-string "(java.util.UUID/randomUUID)"
  {:classes {'java.util.UUID java.util.UUID}})
;;=> #uuid "312ba519-37e2-4109-b164-97fb140b57b0"

To make this work with GraalVM you will also need to add an entry to your reflection config for this class. Also see reflection.json.

By default, SCI only lets you interop with classes explicitly provided in the :classes config. When a method call returns an instance of a class that is not in :classes you won't be able to to interop on that. You can disable this safety measure with {:classes {:allow :all}}.

In JS hosts, to allow interop with anything, use the following config:

{:classes {'js goog/global :allow :all}}


Sci uses a context (internally implemented using an atom) to keep track of state changes like newly defined namespaces and vars. The contents of the context should be considered implementation detail. Every call to eval-string creates a fresh context. To preserve state over multiple evaluations, you can create a context using the same options as those for sci/eval-string.

(def opts {:namespaces {' {'x 1}}})
(def sci-ctx (sci/init opts))

The SCI context can then be re-used over successive invocations of sci/eval-string*:

(sci/eval-string* sci-ctx "") ;;=> 1
(sci/eval-string* sci-ctx "(ns (def x 2) x") ;;=> 2
(sci/eval-string* sci-ctx "") ;;=> 2

In a multi-user environment it can be useful to give each user their own context. This can already be achieved with eval-string, but for performance reasons it may be desirable to initialize a shared context once. This shared context can then be forked for each user so that changes in one user's context aren't visible to other users:

(def forked (sci/fork sci-ctx))
(sci/eval-string* forked "(def forked 1)")
(sci/eval-string* forked "forked") ;;=> 1
(sci/eval-string* sci-ctx "forked") ;;=> Could not resolved symbol: forked

Implementing require and load-file

SCI supports loading code via a hook that is invoked by SCI's implementation of require. The job of this function is to find and return the source code for the requested namespace. This passed-in function will be called with a single argument that is a hashmap with a key :namespace. The value for this key will be the symbol of the requested namespace.

This function should return a map with keys :file (containing the filename to be used in error messages) and :source (containing the source code text). SCI will evaluate that source code to satisfy the call to require. Alternatively the function can return nil which will result in SCI throwing an exception that the namespace could not be found.

The load hook is passed as part of the SCI options via the :load-fn:

(defn load-fn [{:keys [namespace]}]
  (when (= namespace 'foo)
    {:file "foo.clj"
     :source "(ns foo) (def val :foo)"}))
(sci/eval-string "(require '[foo :as fu]) fu/val" {:load-fn load-fn})
;;=> :foo

Note that internally specified namespaces (either the default namespaces that SCI provides itself or those provides via the :namespaces key) will be considered first and if found there, :load-fn will not be called, unless :reload or :reload-all are used:

  "(require '[foo :as fu])
  {:load-fn load-fn
   :namespaces {'foo {'val (sci/new-var 'val :internal)}}})
;;=> :internal

  "(require '[foo :as fu] :reload)
  {:load-fn load-fn
   :namespaces {'foo {'val (sci/new-var 'val :internal)}}})
;;=> :foo

Another option for loading code is to provide an implementation of clojure.core/load-file. An example is presented here.

    (:require [sci.core :as sci]
              [ :as io]))

(spit "example1.clj" "(defn foo [] :foo)")
(spit "example2.clj" "(load-file \"example1.clj\")")

(let [env (atom {})
      opts {:env env}
      load-file (fn [file]
                  (let [file (io/file file)
                        source (slurp file)]
                      {sci/ns @sci/ns
                       sci/file (.getAbsolutePath file)}
                      (sci/eval-string source opts))))
      opts (assoc-in opts [:namespaces 'clojure.core 'load-file] load-file)]
  (sci/eval-string "(load-file \"example2.clj\") (foo)" opts))
;;=> :foo


Implementing a REPL can be done using the following functions:

  • sci/reader: returns reader for parsing source code, either from a string or io/reader
  • sci/parse-next: returns next form from reader
  • sci/eval-form: evaluates form returned by parse-next.

See examples for examples for both Clojure and ClojureScript. Run instructions are included at the bottom of each example.

To include an nREPL server in your sci-based project, you can use babashka.nrepl.


For general information about Clojure and GraalVM, check out clj-graal-docs and graalvm-clojure.

Clojure version

To build native images with GraalVM it is recommended to use clojure 1.10.3 or later.

Use as native shared library

To use SCI as a native shared library from e.g. C, C++, Rust, read this tutorial.


Currently SCI doesn't support deftype and definterface.


Forms evaluated by SCI can produce lazy sequences. In Clojure, dynamic vars and laziness can be a tricky combination and the same goes for dynamic SCI vars.

Consider the following example:

(let [sw     (
      result (sci/binding [sci/out sw] (sci/eval-string "(map print (range 10))"))]
  (println "Output:" (str sw))
  (println "Result:" result))

If the returned lazy seq was realized within the sci/binding scope, the output would be:

Output: 0123456789
Result: (nil nil nil nil nil nil nil nil nil nil)

But because the result is only printed outside of sci/binding the result is:

Execution error (ClassCastException) at (io.cljc:44).
class sci.impl.vars.SciUnbound cannot be cast to class (sci.impl.vars.SciUnbound is in unnamed module of loader clojure.lang.DynamicClassLoader @4c2af006; is in module java.base of loader 'bootstrap')

This happens because by the time the lazy-seq is realized, the binding scope for sci/out is no longer established, and as a result the lazy-seq can no longer be realized (due to the delayed calls to println, a side-effecting call dependends on the value of sci/out, set by sci/binding.

If the result is intented to be serialized as a string, then one could simply serialize while the binding is still in place:

(let [sw (]
  (sci/binding [sci/out sw]
    (let [result (sci/eval-string "(map print (range 10))")]
      (println "Result:" result)
      (println "Output:" (str sw)))))

Note that we moved (println "Result:" result) before (println "Output:" (str sw)), since the first call takes care of realization.


Required: lein, the clojure CLI and GraalVM.

To succesfully run the GraalVM tests, you will have to compile the binary first with script/compile.

To run all tests:


For running individual tests, see the scripts in script/test.



Use clojure -M:bench to benchmark the various phases of sci on the JVM:

$ clojure -M:bench --complete --sexpr "(let [x 1 y 2] (+ x y))" --quick
BENCHMARKING EXPRESSION: (let [x 1 y 2] (+ x y))
-> (let [x 1 y 2] (+ x y))
Evaluation count : 605268 in 6 samples of 100878 calls.
             Execution time mean : 1,105801 µs
    Execution time std-deviation : 209,640200 ns
   Execution time lower quantile : 934,602619 ns ( 2,5%)
   Execution time upper quantile : 1,459172 µs (97,5%)
                   Overhead used : 8,140922 ns

Found 1 outliers in 6 samples (16,6667 %)
	low-severe	 1 (16,6667 %)
 Variance from outliers : 48,1886 % Variance is moderately inflated by outliers
Evaluation count : 82248 in 6 samples of 13708 calls.
             Execution time mean : 8,562699 µs
    Execution time std-deviation : 1,289355 µs
   Execution time lower quantile : 7,557778 µs ( 2,5%)
   Execution time upper quantile : 10,039290 µs (97,5%)
                   Overhead used : 8,140922 ns
-> 3
Evaluation count : 1607688 in 6 samples of 267948 calls.
             Execution time mean : 433,635072 ns
    Execution time std-deviation : 67,003007 ns
   Execution time lower quantile : 378,876890 ns ( 2,5%)
   Execution time upper quantile : 512,818448 ns (97,5%)
                   Overhead used : 8,140922 ns

Use --parse, --evaluate and/or --analyze to bench individual phases (--complete will bench all of them). Leaving out --quick will run criterium/bench instead of criterium/quick-bench.

GraalVM native-image

To benchmark an expression within GraalVM native-image, run script/compile and then run:

$ time ./sci "(loop [val 0 cnt 1000000] (if (pos? cnt) (recur (inc val) (dec cnt)) val))"
./sci    0.92s  user 0.08s system 99% cpu 1.003 total



Copyright © 2019-2021 Michiel Borkent

Distributed under the Eclipse Public License 1.0. This project contains code from Clojure and ClojureScript which are also licensed under the EPL 1.0. See LICENSE.

Can you improve this documentation? These fine people already did:
Michiel Borkent, Lee Read, sogaiu, Nate Jones, Tienson Qin, Crispin Wellington, Tommi Reiman, Maurício Szabo, Kamil Toman, Kim Kinnear, Jeroen van Dijk, Peter Monks, Sam Ritchie, Rahuλ Dé & JC
Edit on GitHub

cljdoc is a website building & hosting documentation for Clojure/Script libraries

× close