diff --git a/README.md b/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..8dd3711a33661be4280e0218795a41775d7bf56a
--- /dev/null
+++ b/README.md
@@ -0,0 +1,121 @@
+# ASCII-style alignment pileup
+
+## Description
+
+Generates an ASCII-style pileup of read alignments in one or more BAM files
+against one or more regions specified in a BED file.
+
+## Usage
+
+```sh
+ascii_alignment_pileup.R [-hv] [OPTIONS] bed bam [bam2 ...]
+```
+
+## Requirements
+
+* R 3.6.0
+* GenomicAlignments 1.20.0
+* optparse 1.6.2
+* rtracklayer 1.44.0
+* tools 3.6.0
+
+## Input files
+
+* [BED](https://www.ensembl.org/info/website/upload/bed.html) file
+* [BAM](https://samtools.github.io/hts-specs/) file(s)
+* Optional: [FASTA](https://en.wikipedia.org/wiki/FASTA_format) file compressed
+  with [`bgzip`](http://www.htslib.org/doc/bgzip.html)
+* Optional:
+  [GFF/GTF/GFF3](https://en.wikipedia.org/wiki/General_feature_format) file
+
+## Output files
+
+* Custom file format:
+
+```console
+>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>        hsa-let-7a-1
+.....>>>>>>>>>>>>>>>>>>>>>>.....................................................        hsa-let-7a-5p
+........................................................>>>>>>>>>>>>>>>>>>>>>...        hsa-let-7a-3p
+TGGGATGAGGTAGTAGGTTGTATAGTTTTAGGGTCACACCCACCACTGGGAGATAACTATACAATCTACTGTCTTTCCTA        9:94175957-94176036:+
+TACCATGAGGTAGTAGGTTGTATAGTT.....................................................        1
+...CATGAGGTAGTAGGTTGTATAGTT.....................................................        10
+...GA-GAGGTAGTAGGTTGTATAGTT.....................................................        2
+....A-GAGGTAGTAGGTTGTATAGTT.....................................................        19
+.....TGAGGTAGTAGGTTGTATAGTT.....................................................        17
+......GAGGTAGTAGGTTGTATAGTT.....................................................        33
+.......AGGTAGTAGGTTGTATAGTT.....................................................        9
+.......AGGTAGTAGGTTGTATAGTTT....................................................        2
+........GGTAGTAGGTTGTATAGTT.....................................................        7
+...................................................GATAACTATACAATCTACTGTCTT.....        1
+......................................................AACTATACAATCTACT..........        1
+........................................................CTATACAATCTACTGTCTTTCT..        28
+........................................................CTATACAATCTACTGTCTTTC-T.        22
+........................................................CTATACAATCTACTGTCTTTCC..        19
+........................................................CTATACAATCTACTGTCTTTC...        12
+........................................................CTATACAATCTACTGTCTTTCTT.        2
+........................................................CTATACAATCTACTGTC.......        1
+........................................................CTATACAATCTACTGTCTT.....        1
+........................................................CTATACAATCTACTGTCTTTCG..        1
+.........................................................TATACAATCTACTGTCTTTCT..        4
+.........................................................TATACAATCTACTGTCTTTC-T.        4
+.........................................................TATACAATCTACTGTCTTTC...        2
+.........................................................TATACAATCTACTGTCTTTCC..        1
+.........................................................TATACAATCTACTGTCTTTCCT.        1
+```
+
+## Options
+
+```console
+--reference=FILE
+        Reference genome sequence in FASTA format. The file *MUST* be compressed
+    with BGZIP. If supplied, the reference sequence for the query region(s) will
+    be added to the output. Note that on the first run with a specific reference
+    genome file, an FAI index is generated which will take some time.
+
+--annotations=FILE
+        Annotation file in GFF/GTF format used to annotate sequences. If
+    supplied, features overlapping the query region(s) will be visualized in the
+    output. Ensure that the argument to option `annotation-name-field`
+    corresponds to a field in the annotations, otherwise the script will fail.
+
+--output-directory=DIR
+        Output directory. One output file will be created for each region in
+    `--bed` and the filenames will be generated from the basenames of the
+    supplied BAM file(s) and the name field (4th column) of the BED file.
+    [default "."]
+
+--maximum-region-width=INT
+        Maximum input region width. Use with care as wide regions will use
+    excessive resources. [default 200]
+
+--do-not-collapse-alignments
+        Show alignments of reads with identical sequences individually.
+
+--minimum-count=INT
+        Alignments of reads with less copies than the specified number will not
+    be printed. Option is not considered if `do-not-collapse-alignments` is
+    set. [default 1]
+
+--annotation-name-field=STR
+        Annotation field used to populate the `name` column in the output.
+    [default "Name"]
+
+--padding-character=CHAR
+        Character used for padding alignments. [default "."]
+
+--indel-character=CHAR
+        Character to denote insertions and deletions in alignments.
+    [default "-"]
+
+-h, --help
+        Show this information and die.
+
+-v, --verbose
+        Print log messages to STDOUT.
+```
+
+## Contact
+
+Email: <zavolab-biozentrum@unibas.ch>
+
+&copy; 2019 Zavolab, Biozentrum, University of Basel