Install with Pen
Pen.el on GitHub https://github.com/semiosis/pen.el/
Tutorial https://mullikine.github.io/posts/pen-el-tutorial/

Summary

I have put together an LSP server that utilises large LMs, so that interacting with your computer whilst utilising AI systems is fully transparent and controllable.

You have control over:

  • What information is sent to the language model (could be to Google, Microsoft, etc.).
    • For example, a prompt may watch your screen and send that information to the LM. Pen.el gives you control of that information, and you may also, for example, generate fake information, or obscure information by placing a sed command, for example, in the prompt preprocessor.
  • How much information is sent
  • Which language models you will use
  • Building software language model agnostically
    • Design for multiple models and run prompts on your own private models, if you want

Only this way can you avoid trusting large corporations with all of your private information. Make no mistake, almost all (effectively all) information is about to be effectively harvested due to language models and inference, while using a computer. It’s only out of good will that this doesn’t happen. But how long will that last now the LM technology cat is out of the bag?

It’s therefore imperative to be able to control this flow of information. You need a language server for large LMs, to be your Copilot, your personal assistant, your Google search, Grammerly, etc., an AGI interface that you have full control over. That is part of what Pen.el is. It’s an LSP server for anything.

If you understand what emacs is then you’re probably aware of the way that people try to do all of their computing through emacs. And that is one reason for emacs being the ideal choice for an LSP server. The other big reason is that everything you do through emacs can fit within an GPT-3 prompt.

Current state of Pen.el LSP server

It is very nearly ready to be used for all emacs modes, and for other editors. The issue currently is that LSP servers require access to a real file on the same filesystem as the LSP server is running. Currently your software would have to be running inside the docker container for the lsp server to work with it. You can, for instance, install VSCode into the container and connect. However, I’m working on a solution to have contact with the external world. That involves possibly using a bind/sshfs mount.

Prompting server

The Pen.el container also functions as a prompting server.

Pen.el may very well be simultaneously the first open-source prompting server and LSP server for prompting.

See this blog article for more on how it can be used in your own pipelines. It’s dead simple, just pipe into the penf script along with the prompt function name and arguments the same way the efm-langserver does it shown below.

Something like this

1

pena -nj --pool --temp 0.9 pf-very-witty-pick-up-lines-for-a-topic/1 imagination

See https://mullikine.github.io/posts/pen-el-host-interop-and-client-server/

Usage

Firstly install Pen.el. The best way to use it currently is via the docker container.

Installation instructions
https://semiosis.github.io/pen/

Configuration

The LSP configuration lives here
$HOME/.pen/efm-langserver-config.yaml

Under the list of languages recognised there is one called global.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10

languages:
  text:
    - <<: *pen-world-language
    - <<: *pen-world-language-completion
    - <<: *glossary1

  global:
    - <<: *pen-world-language
    - <<: *pen-world-language-completion
    - <<: *glossary1

As you can see I have added glossary to global. This enables hover documentation for where your cursor is at.

The prompt function pf-define-word-for-glossary/1 will be used.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11

glossary1: &glossary1
  # The impementation is a combination of documentation and hover
  # documentation is provided to the thing under the cursor

  # The rest go to the hover provider as signature docs
  hover-command: "pen-ci -f -a penf -u -nj --pool pf-define-word-for-glossary/1"
  # hover-command: "timeout -k 0.1 -s KILL 0.1 pen-ci -f -a cat"
  hover-stdin: true
  # hover-type: "markdown"
  # hover-type: "plaintext"
  format-command: 'pen-pretty-paragraph'

Within the pen.el repository, you have a list of modes defined here.

This is for the emacs client when running emacs inside the docker container, which you can try out for yourself currently.

http://github.com/semiosis/pen.el/blob/master/src/pen-lsp-client.el

1
2
3
4
5
6

(defset pen-lsp-modes '(text-mode
                        emacs-lisp-mode
                        sh-mode
                        org-mode awk-mode eww-mode
                        special-mode python-mode
                        prog-mode))

However, when completed this wont need to be configured for use with other editors.

How it works

pen.el is a prompting pipeline which utilises templating and many different language and world models to generate responses. That could be used to generate code, guitar tabs, recipes, to imagine a programming interface or to interact with a chatbot, for example.

pen.el is written in emacs lisp, and is daemonised to server an LSP server, which in turn is used by another emacs or to different editors to provide documentation and refactoring tools.

Demo

Customizing the prompts

That is done via the following repository:

https://github.com/semiosis/prompts/tree/master/prompts

Example debug information

For your interest.

The command M-x pen-diagnostics-show-context will display information on the last prompt.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86

ai21-last-output: ''
ai21-last-output-fp: $HOME/.pen/temp/ai21-temp.txt
aix-last-output: ''
aix-last-output-fp: $HOME/.pen/temp/aix-temp.txt
alephalpha-last-output: ''
alephalpha-last-output-fp: $HOME/.pen/temp/alephalpha-temp.txt
hf-last-output: ''
hf-last-output-fp: $HOME/.pen/temp/hf-temp.txt
human-last-output: 'Once upona time

  ###'
human-last-output-fp: $HOME/.pen/temp/human-temp.txt
last-final-command: ALSO_EXPORT="" PEN_PROMPT="Glossary of terms.\n###\nossified\nDefinition<pen-colon>
  Turn into bone or bony tissue.\n###\n18\nDefinition<pen-colon>" PEN_LM_COMMAND="openai-complete.sh"
  PEN_MODEL="davinci-codex" PEN_WHITESPACE_SUPPORT="" PEN_ENGINE="OpenAI Codex" PEN_API_ENDPOINT="https://api.openai.com"
  PEN_PAYLOADS="" PEN_QUERY="" PEN_COUNTERQUERY="" PEN_LOGPROBS="10" PEN_APPROXIMATE_PROMPT_LENGTH=42
  PEN_ENGINE_MIN_TOKENS=0 PEN_ENGINE_MAX_TOKENS=2049 PEN_MIN_TOKENS=0 PEN_MAX_TOKENS=122
  PEN_REPETITION_PENALTY="" PEN_FREQUENCY_PENALTY="" PEN_PRESENCE_PENALTY="" PEN_LENGTH_PENALTY=""
  PEN_MIN_GENERATED_TOKENS=3 PEN_MAX_GENERATED_TOKENS=80 PEN_TEMPERATURE="0.5" PEN_MODE=""
  PEN_STOP_SEQUENCE="##" PEN_STOP_SEQUENCES="[\"##\",\"<delim-1>\",\"<delim-1>\"]"
  PEN_DOCUMENTS= PEN_TOP_P="1" PEN_TOP_K=1 PEN_FLAGS= PEN_CACHE= PEN_USER_AGENT="emacs/pen"
  PEN_TRAILING_WHITESPACE=" " PEN_N_COMPLETIONS="1" PEN_ENGINE_MIN_GENERATED_TOKENS=3
  PEN_ENGINE_MAX_GENERATED_TOKENS=4096 PEN_COLLECT_FROM_POS=94 PEN_END_POS=94 PEN_N_JOBS="1"
  PEN_SEARCH_THRESHOLD= PEN_GEN_UUID="d39816c8-9c67-4af0-b858-263ac7e55d2a" PEN_GEN_TIME=1640780265.3621588
  PEN_GEN_DIR="/root/.pen/results/results_1640780265.3621588_29.12.21_d39816c8-9c67-4af0-b858-263ac7e55d2a"
  PEN_INJECT_GEN_START="" lm-complete
last-final-prompt: 'Glossary of terms.

  ###

  ossified

  Definition: Turn into bone or bony tissue.

  ###

  18

  Definition: <END>'
last-pen-command: '"penf" "-u" "pf-define-word-for-glossary/1" "18" "nil" "nil"'
last-pen-command-exprs: '"penf" "-u" "pf-define-word-for-glossary/1" "18"'
last-prompt-data: null
lm-complete-results: ''
lm-complete-stderr: ''
lm-complete-stdout: '/tmp/results_29.12.21__1640780269_75be3eca-250b-41a2-b6b9-25c65a5660f6_DKHR1

  8ed897a_U66VG'
nlpcloud-last-output: ''
nlpcloud-last-output-fp: $HOME/.pen/temp/nlpcloud-temp.txt
offline-last-output: 'Glossary of terms.

  ###

  ossified

  Definition: Turn into bone or bony tissue.

  ###

  something

  Definition: PEN MODEL DummyModel prompt Glossary of terms n nossified nDefinition
  Turn into bone or bony tissue'
offline-last-output-fp: $HOME/.pen/temp/offline-temp.txt
openai-last-output: 'Glossary of terms.

  ###

  ossified

  Definition: Turn into bone or bony tissue.

  ###

  )/

  Definition: A word used to replace a word that cannot be used by itself but can
  be used with another word.

  '
openai-last-output-fp: $HOME/.pen/temp/openai-temp.txt
pen-force-engine: Offline
pen-prompt-functions-failed: null
pen-prompts-failed: null
ruby-gen-next-user-prompt: $HOME/.pen/temp/ruby-gen-next-user-prompt.txt
semantic-path: spacemacs-buffer

Also, for your interest.

The command M-x pen-customize will start the following control panel.

  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149

For help using this buffer, see [Easy Customization] in the [Emacs manual].

Operate on all settings in this buffer:
[ Revert... ] [ Apply ] [ Apply and Save ]


pen group:  Group definition missing.
      [ State ]: something in this group has been changed outside customize.

Hide iλ-thin: Boolean: [Toggle]  on (non-nil)
   [ State ]: SAVED and set.
   thin-client mode toggle

Hide pen-avoid-divulging: Boolean: [Toggle]  off (nil)
   [ State ]: CHANGED outside Customize.
   Avoid divulging information

Show Value pen-completers-directory
   Directory where personal .completer files are located

Show Value pen-contrib-directory
   Personal pen-contrib.el respository

Show Value pen-cost-efficient
   Avoid spending money

Hide pen-debug: Boolean: [Toggle]  on (non-nil)
   [ State ]: CHANGED outside Customize.
   When debug is on, try is disabled, and all errors throw an exception

Hide pen-default-engine: nil
   [ State ]: CHANGED outside Customize. (mismatch)
   Default engine

Show Value pen-default-lm-command
   Default LM completer script

Hide pen-default-logprobs: Integer: 10
   [ State ]: CHANGED outside Customize.
   The default logprobs value

Show Value pen-describe-images
   Describe images

Show Value pen-engines-directory
   Personal engine respository

Hide pen-eww-text-only: Boolean: [Toggle]  off (nil)
   [ State ]: CHANGED outside Customize.
   eww text mode only

Show Value pen-fav-programming-language
   By setting pen-fav-programming-language, you set a default language to translate into. More

Show Value pen-fav-world-language
   By setting pen-fav-world-language, you set a default language to translate into. More

Hide pen-force-engine: String: Offline
   [ State ]: CHANGED outside Customize.
   Force using this engine

Show Value pen-force-few-completions
   Forcing only few completions will speed up Pen.el, but not by much usually

Show Value pen-force-logprobs
   The logprobs to force

Show Value pen-force-n-collate

Show Value pen-force-n-completions

Show Value pen-force-no-uniq-results

Show Value pen-force-one

Show Value pen-force-single-collation
   Forcing only one collation will speed up Pen.el

Hide pen-force-strip-unicode: Boolean: [Toggle]  off (nil)
   [ State ]: CHANGED outside Customize.
   Strip unicode from input

Show Value pen-force-temperature
   The temperature to force

Show Value pen-glossaries-directory
   Personal glossary respository

Show Value pen-ink-disabled
   Disable ink. Useful if it’s breaking

Show Value pen-libre-only
   Only use libre engines

Hide pen-logprobs-on: Boolean: [Toggle]  on (non-nil)
   [ State ]: CHANGED outside Customize.
   Boolean to enable/disable logprobs

Hide pen-memo-prefix: String: mele-host
   [ State ]: CHANGED outside Customize.
   memoize file prefix

Hide pen-n-simultaneous-requests: Integer: 5
   [ State ]: CHANGED outside Customize.
   The number of requests to make in parallel using lm-complete

Show Value pen-nlsh-histdir
   Directory where history files for nlsh

Hide pen-obtain-probabilities: Boolean: [Toggle]  off (nil)
   [ State ]: CHANGED outside Customize.
   Also query for the probabilities, when prompting

Show Value pen-override-lm-command
   Override LM completer script

Show Value pen-penel-directory
   Personal pen.el respository

Show Value pen-preview-token-length
   The number of tokens to generate in order to get a preview of what to further generate

Show Value pen-prompt-discovery-recursion-depth
   The number of git repositories deep that pen.el will go looking

Hide pen-prompt-force-engine-disabled: Boolean: [Toggle]  on (non-nil)
   [ State ]: CHANGED outside Customize.
   Do not allow prompts to override the engine override

Show Value pen-prompt-function-prefix
   Prefix string to prepend to prompt function names

Show Value pen-prompts-directory
   Personal prompt respository

Show Value pen-prompts-library-directory
   The directory where prompts repositories are stored

Show Value pen-sh-update
   Export UPDATE=y when executing sn and such

Show Value pen-snippets-directory
   Personal snippets respository

Show Value pen-term-cl-refresh-after-fz
   While in term, send a C-l just after selecting for insertion

Show Value pen-user-agent
   User Agent for self identification