OSR-AI

AI service prompter

Installation

1. npm i -g @plastichub/osr-ai
2. create a file with your OpenAI key, in C:\\Users\\TeleTubbie\\.config\\osr\\config.json, with the following content

{
    "openai": {
        "see":"https://platform.openai.com/api-keys"
        "key": "some katbotisch key"
    }
}

CLI parameters

• --src : the path to a source file
• --files : for binary files as images, please use this instead
• --dst : the path to the output file (single quotes!), defaults to ${SRC_DIR}/${SRC_NAME}-output.md
• --logLevel : warn|info|trace|debug|error|fatal, defaults to info
• --filters : a list of filters to be applied Markdown JSONUnescape JSONPretty AlphaSort code JSONParse trim defaults to code (OpenAI returns Markdown escaped code sequences)
• --cache : enable cache, when enabled (default), it skips the request when the destination file already exists
• --model : OpenAI model, defaults to chatgpt-4o-latest, see official documentation or run osr-ai chatgpt --help

Example, describe the content of a file

osr-ai chatgpt prompt --source="./tests/chatgpt/pipe.txt" --query="whats in here, add a link to their homepage, check it and add a summary of their activities, as markdown table"

Or

osr-ai chatgpt prompt --source=".tests/chatgpt/pipe.txt" --query="./query.md"

whereby ./query.md contains the query : whats in here, add a link to their homepage, check it and add a summary of their activities, as markdown table

Or

cat tests/chatgpt/pipe.txt | osr-ai chatgpt prompt  --query="whats in here, add a link to their homepage, check it and add a summary of their activities, as markdown table"

whereby pipe.txt contains

abb
bigtree
cead
cidepa
kuga
motovario
nagami
siemens
teknic

Output

| Company | Homepage | Summary | |----------|-----------------------------------------------|---------------------------------------------------------| | abb | Homepage | ABB is a leading global technology company. They focus on electrification, robotics, and automation solutions. | | bigtree | Homepage | Bigtree is a software development company specializing in web and mobile applications. | | cead | Homepage | CEAD is a company that specializes in 3D printing and additive manufacturing solutions. | | cidepa | Homepage | Cidepa is a manufacturer of industrial machinery and equipment. | | kuga | Homepage | Kuga is an automotive company that designs and manufactures electric vehicles. | | motovario | Homepage | Motovario is a global manufacturer of power transmission and motion control solutions. | | nagami | Homepage | Nagami is a design studio specializing in computational design and digital fabrication. | | siemens | Homepage | Siemens is a multinational conglomerate that provides a wide range of products and services in various industries. | | teknic | Homepage | Teknic is a manufacturer of high-performance motion control products. |

Example, describe the content of an image (Example)

osr-ai chatgpt prompt --files='./tests/chatgpt/files/2.jpg' --query="whats in here, as json" --dst='&{FILE_DIR}/&{FILE_NAME}-content.json`

output

{
  "image": {
    "people": [
      {
        "handle": "@judi_wakhungu",
        "position": "left",
        "description": "A person wearing an orange shirt, a hat, and glasses, holding a phone."
      },
      {
        "handle": null,
        "position": "center",
        "description": "A person standing in the doorway, wearing a white shirt and a cap."  
      },
      {
        "handle": "@katatungo",
        "position": "right",
        "description": "A person wearing a white t-shirt with text, and black pants."        
      }
    ],
    "textOverlay": "This door is not painted...",
    "door": {
      "description": "A rustic, two-part wooden door with blue and teal hues and a gold latch."
    },
    "tags": [
      "@katatungo",
      "@judi_wakhungu"
    ]
  }
}

Example, transform code

osr-ai chatgpt prompt --source="./tests/chatgpt/constants.ts" --query="rewrite this enum as object literal" --model=gpt-4 --debug

whereby ./tests/chatgpt/constants.ts contains

export enum SpainCities {
    Madrid = "40.4168,-3.7038,12",
    Barcelona = "41.3851,2.1734,12",
    Valencia = "39.4699,-0.3763,12",
    Seville = "37.3891,-5.9845,12",
    Malaga = "36.7213,-4.4214,12",
    Bilbao = "43.2630,-2.9340,12",
    Alicante = "38.3452,-0.4810,12",
    Zaragoza = "41.6488,-0.8891,12",
    Palma = "39.5696,2.6502,12",
    Murcia = "37.9922,-1.1307,12",
    Granada = "37.1773,-3.5986,12",
    Cordoba = "37.8882,-4.7794,12",
    Valladolid = "41.6520,-4.7286,12",
    Vigo = "42.2406,-8.7207,12",
    Gijon = "43.5322,-5.6611,12",
    Lleida = "41.6176,0.6200,12",
    Tarragona = "41.1189,1.2445,12",
    Cadiz = "36.5297,-6.2924,12",
    Almeria = "36.8340,-2.4637,12",
    Oviedo = "43.3614,-5.8593,12",
    Badajoz = "38.8794,-6.9706,12",
    Santander = "43.4623,-3.8090,12",
    Logrono = "42.4627,-2.4441,12",
    Salamanca = "40.9701,-5.6635,12",
    Pamplona = "42.8125,-1.6458,12",
}

Output

export const BarcelonaTowns = {
    Barcelona: "41.3850639,2.1734035",
    Badalona: "41.453966,2.24487",
    "L'HospitaletDeLlobregat": "41.3618046,2.0977132",
    Terrassa: "41.5654824,2.0109316",
    Sabadell: "41.5441063,2.1068342",
    Mataro: "41.540977,2.4440367",
    SantaColomaDeGramenet: "41.4679772,2.2081475",
    CornellaDeLlobregat: "41.3666298,2.0971758",
    SantCugatDelValles: "41.4888889,2.1019444",
    Castelldefels: "41.2755556,1.995",
    CerdanyolaDelValles: "41.4913889,2.1169444",
    Granollers: "41.6072222,2.2686111",
    VilanovaIGeltru: "41.2155556,1.7383333",
    Viladecans: "41.3163889,2.0052778",
    ElPratDeLlobregat: "41.2971628,2.0700942",
    MolletDelValles: "41.5272964,2.1870736",
    Gava: "41.3166667,2.0666667",
    SantBoiDeLlobregat: "41.3669444,2.0569444",
}

Example, transcribe an audio file, to text

osr-ai chatgpt transcribe --debug --source="./tests/chatgpt/real.mp3" --dst="./tests/chatgpt/real.txt"

Example, create code for SCAD

osr-ai chatgpt prompt --query="create a scene, with all possible primitives, lined up along one line, for SCAD" --model=gpt-4 --dst="./tests/chatgpt/primitives.scad" --system="./tests/chatgpt/scad-system.json"

whereby ./tests/chatgpt/scad-system.json refers to the OpenAI assistant instructions, helping to reuse optimizations and rules:

[
    {
        "role": "system",
        "content": "don't add footer comments"
    },
    {
        "role": "system",
        "content": "don't wrap it in markdown codeblocks"
    },
    {
        "role": "system",
        "content": "don't add header or footer notes" 
    },
    {
        "role": "system",
        "content": "comment all SCAD functions" 
    }  
]

Output (works perfectly :)

//Create a cube with side length 10 at origin
translate([-35, 0, 0])  
cube(10, center = true);  

//Create a sphere with diameter 10 at coordinate (20,0,0)
translate([-15, 0, 0])  
sphere(5); 

//Create a cylinder with height 10, lower diameter 10, upper diameter 5 at coordinate (40,0,0)
translate([5, 0, 0])  
cylinder(h = 10, r1 = 5, r2 = 2.5, center = true);   

//Create a polyhedron with vertices at (0,0,0), (10,0,0), (0,10,0), and (0,0,10) at coordinate (60,0,0)
translate([25, 0, 0]) 
polyhedron(points=[[-5,-5,-5],[5,-5,-5],[-5,5,-5],[0,0,5]], faces=[[0,1,2],[0,2,3],[0,3,1],[1,3,2]]); 

Example, summarize the code from above

osr-ai chatgpt prompt --source="./tests/chatgpt/primitives.scad" --model=gpt-4 --dst="./tests/chatgpt/primitives-next.scad" --system="./tests/chatgpt/scad-system.json" --query="whats this code doing, it's for OpenSCAD"

Output

This OpenSCAD code does the following:

- Creates a cube with a side length of 10 at the origin, translated from the origin by -35 units along the X-axis. The cube is centered at this new location. 

- Creates a sphere with a diameter of 10 units, translated from the origin by -15 units along the X-axis.

- Creates a cylinder with a height of 10 units, a lower diameter of 10 units, and an upper diameter of 5 units. This is translated from the origin by 5 units along the X-axis, and the cylinder is centered at its location.

- Creates a polyhedron with vertices defined by the points (0,0,0), (10,0,0), (0,10,0), and (0,0,10). It is translated from the origin by 25 units along the X-axis.

In summary, the code creates a sequence of four different 3D geometric forms (a cube, a sphere, a cylinder, and a polyhedron). Each shape is positioned in a line along the X-axis with specifically defined spaces between them.

Example, convert C code to Javascript

osr-ai chatgpt prompt --source="./tests/chatgpt/math_statistics.h" --model=gpt-4 --dst="./tests/chatgpt/math_statistics.js" --system="./tests/chatgpt/scad-system.json" --query="Write it as Javascript"

whereby math_statistics.h is

// C++11 solution that is standards compliant. Return type is deduced automatically
template <class L, class R>
static inline constexpr auto MIN(const L lhs, const R rhs) -> decltype(lhs + rhs)
{
  return lhs < rhs ? lhs : rhs;
}
template <class L, class R>
static inline constexpr auto MAX(const L lhs, const R rhs) -> decltype(lhs + rhs)
{
  return lhs > rhs ? lhs : rhs;
}
template <class T>
static inline constexpr const T ABS(const T v)
{
  return v >= 0 ? v : -v;
}
#else
// Using GCC extensions, but Travis GCC version does not like it and gives
//  "error: statement-expressions are not allowed outside functions nor in template-argument lists"
#define MIN(a, b) \
  ({__typeof__(a) _a = (a); \
      __typeof__(b) _b = (b); \
      _a < _b ? _a : _b; })

#define MAX(a, b) \
  ({__typeof__(a) _a = (a); \
      __typeof__(b) _b = (b); \
      _a > _b ? _a : _b; })

#define ABS(a) \
  ({__typeof__(a) _a = (a); \
      _a >= 0 ? _a : -_a; })

#endif

class Statistic
{
public:
  Statistic();  // "switches on/off" stdev run time
  void clear(); // "switches on/off" stdev run time
  void add(const float);

  // returns the number of values added
  uint32_t count() const { return _cnt; }; // zero if empty
  float sum() const { return _sum; };      // zero if empty
  float minimum() const { return _min; };  // zero if empty
  float maximum() const { return _max; };  // zero if empty
  float average() const;                   // NAN if empty
  float mean() const;                      // zero if empty

#ifdef STAT_USE_STDEV
  float variance() const;       // NAN if empty
  float pop_stdev() const;      // population stdev  // NAN if empty
  float unbiased_stdev() const; // NAN if empty
#endif

protected:
  uint32_t _cnt;
  float _sum;
  float _min;
  float _max;
#ifdef STAT_USE_STDEV
  float _ssqdif; // sum of squares difference
#endif
};

Output

class Statistics {
  constructor() {
    this._cnt = 0;
    this._sum = 0;
    this._min = null;
    this._max = null;
    this._ssqdif = 0;
  }

  clear() {
    this._cnt = 0;
    this._sum = 0;
    this._min = null;
    this._max = null;
    this._ssqdif = 0;
  }

  add(value) {
    if (typeof this._min === 'undefined' || this._min > value) this._min = value;
    if (typeof this._max === 'undefined' || this._max < value) this._max = value;
    this._sum += value;
    this._cnt += 1;
    this._ssqdif += value * value; // for variance and stdev
  }

  count() {
    return this._cnt;
  }

  sum() {
    return this._sum;
  }

  minimum() {
    return this._min;
  }

  maximum() {
    return this._max;
  }

  average() {
    return this._cnt ? (this._sum / this._cnt) : NaN;
  }

  mean() {
    return this.average();
  }

  variance() {
    if (!this._cnt) return NaN;
    let mean = this.average();
    return (this._ssqdif - 2 * mean * this._sum + this._cnt * mean * mean) / this._cnt;
  }

  pop_stdev() {
    return Math.sqrt(this.variance());
  }

  unbiased_stdev() {
    if (this._cnt < 2) return NaN;
    let mean = this.average();
    return Math.sqrt((this._ssqdif - 2 * mean * this._sum + this._cnt * mean * mean) / (this._cnt - 1));
  }
}

function min(lhs, rhs) {
  return lhs < rhs ? lhs : rhs;
}
function max(lhs, rhs) {
  return lhs > rhs ? lhs : rhs;
}
function abs(v) {
  return v >= 0 ? v : -v;
}

Example, lets get cozy, find some hotels on a lonely island, for a lonely rider :)

osr-ai chatgpt prompt --model=gpt-4 --dst="./tests/chatgpt/hotels.md" --query="Create a list of hotels, on Lamu, Kenya. Create a markdown bullet list, with price, link, and a link on GoogleMaps - add average rating"

Output

• Price: $200 per night
  • Booking Link
  • Google Maps Location
  • Average Rating: 4.5/5

• Majlis Resort

  • Price: $350 per night
  • Booking Link
  • Google Maps Location
  • Average Rating: 4.7/5

• Forodhani House

  • Price: $150 per night
  • Booking Link
  • Google Maps Location
  • Average Rating: 4.2/5

• Baytil Ajaib

  • Price: $100 per night
  • Booking Link
  • Google Maps Location
  • Average Rating: 4.8/5

• Petley's Inn

  • Price: $50 per night
  • Booking Link
  • Google Maps Location
  • Average Rating: 3.7/5

Todos

ChatGPT

• [x] model select
• [ ] loaders
• [ ] language system defaults
• [ ] output templates
• [ ] auto-functions, eg "create file , ... "
• [ ] support Urls, for source, and stdin

OSR Todos

• [ ] Loader (mime/glob) -> storage (mem, mongo,...) -> query -> cache
• [ ] session cache (chat history replay)

References

API Shit

• JSON Schema to TS

OpenAI

• Docs

Google

• Inpaint

Google Gemini

• Intro Model Tuning
• Function Calling

LLM Tooling

• langchain - loaders

Misc

• Voice control for VScode

Images

• ml-mgie (modifier)