ApiForge

AI-powered .NET library that reverse-engineers API integrations from recorded browser traffic (HAR files). Works with local LLMs (LM Studio) or any compatible AI endpoint.

How It Works

1. Record — Playwright opens a browser, you perform the action, traffic is saved as HAR + cookies
2. Analyze — LLM agent identifies the target request, traces dynamic values (tokens, session IDs) back through prior responses, and builds a dependency DAG
3. Generate — Optionally produces runnable integration code from the DAG

Quick Start

dotnet build

Record

var client = new ApiForgeClient();
await client.RecordHarAsync("traffic.har", "cookies.json");

Analyze

// Local AI (LM Studio)
var client = new ApiForgeClient(
    apiKey: "lm-studio",
    model: "openai/gpt-oss-20b",
    endpoint: "http://127.0.0.1:1234/v1"
);

var result = await client.AnalyzeAsync(
    prompt: "send a message",
    harFilePath: "traffic.har",
    cookiePath: "cookies.json",
    generateCode: true,
    maxSteps: 25
);

Environment Variables

Variable	Description
AI_API_KEY	API key (or lm-studio for local)
AI_BASE_URL	Endpoint (e.g. http://127.0.0.1:1234/v1)
AI_MODEL	Model name

Dependencies

• .NET 8.0
• Microsoft.Playwright — Browser recording
• AngleSharp — HTML parsing (used by demo)

SchoolSoftDemo

Included demo app that uses ApiForge to interact with SchoolSoft (Swedish school platform):

• Headless SSO login via GrandID (username + password + SMS OTP)
• Session persistence — SSO cookies (JSESSIONID, Shibboleth session) are saved to session_tokens.json and reused across runs. Sessions are validated both locally (age check) and online before reuse
• Device cookie — GrandID device cookie saved to grandid_device.json after first SMS verification, skipping 2FA on subsequent logins
• Authenticated HttpClient — TokenManager.CreateAuthenticatedClient() builds an HttpClient with all stored cookies injected, ready to call any SchoolSoft endpoint
• Schedule fetching — weekly timetable with day/time/subject/room/group
• Attendance — student roster and attendance status per lesson
• HAR recording — record new flows with the authenticated session and analyze with local LLM

Example: Fetch students for a lesson

=== Schema ===
Ange vecka (Enter = v9, 'q' = avsluta):

Hittade 16 lektioner:

#    Dag        Tid          Amne                 Sal              Grupp
------------------------------------------------------------------------------------
1    Monday     8:30-9:50    Programmering 1      D215             TE22B
2    Monday     10:10-11:30  Webbutveckling 1     D215             TE22A
3    Tuesday    8:30-9:50    Teknik 1             D215             TE22B
...

Valj lektion (1-16) for narvaro, eller Enter for ny vecka: 1

Hamtar narvaro for Programmering 1 (8:30-9:50)...

=== Programmering 1 — Mandag 8:30 (80 min) ===
    Status: Ej genomford
    Elever: 11

  Namn                           Status
  --------------------------------------------------
  Andersson Erik                 Narvarande
  Berg Sofia                     Narvarande
  Johansson Liam                 Franvarande
  Lindgren Maja                  Sen ankomst
  ...

License

MIT