Module: PWN::AI::Ollama

Defined in:
lib/pwn/ai/ollama.rb

Overview

This plugin is used for interacting w/ Ollama's REST API using the 'rest' browser type of PWN::Plugins::TransparentBrowser. This is based on the following Ollama API Specification: https://api.openai.com/v1

Class Method Summary collapse

Class Method Details

.authors ⇒ Object

Author(s)

0day Inc. support@0dayinc.com



577
578
579
580
581
 
# File 'lib/pwn/ai/ollama.rb', line 577

public_class_method def self.authors
  "AUTHOR(S):
    0day Inc. <support@0dayinc.com>
  "
end

.chat(opts = {}) ⇒ Object

Supported Method Parameters

response = PWN::AI::Ollama.chat( request: 'required - message to Ollama' model: 'optional - model to use for text generation (defaults to PWN::Env[:ollama][:model])', temp: 'optional - creative response float (deafults to PWN::Env[:ollama][:temp])', system_role_content: 'optional - context to set up the model behavior for conversation (Default: PWN::Env[:ollama][:system_role_content])', response_history: 'optional - pass response back in to have a conversation', speak_answer: 'optional speak answer using PWN::Plugins::Voice.text_to_speech (Default: nil)', timeout: 'optional timeout in seconds (defaults to 900)', spinner: 'optional - display spinner (defaults to false)' )



493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
 
# File 'lib/pwn/ai/ollama.rb', line 493

public_class_method def self.chat(opts = {})
  engine = PWN::Env[:ai][:ollama]
  request = opts[:request]
  max_prompt_length = engine[:max_prompt_length] ||= 1_000_000
  request_trunc_idx = ((max_prompt_length - 1) / 3.36).floor
  request = request[0..request_trunc_idx]

  model = opts[:model] ||= engine[:model]
  raise 'ERROR: Model is required.  Call #get_models method for details' if model.nil?

  temp = opts[:temp].to_f ||= engine[:temp].to_f
  temp = 1 if temp.zero?

  rest_call = 'ollama/v1/chat/completions'

  response_history = opts[:response_history]

  max_tokens = response_history[:usage][:total_tokens] unless response_history.nil?

  system_role_content = opts[:system_role_content] ||= engine[:system_role_content]

  system_role = {
    role: 'system',
    content: system_role_content
  }

  user_role = {
    role: 'user',
    content: request
  }

  response_history ||= { choices: [system_role] }
  choices_len = response_history[:choices].length

  http_body = {
    model: model,
    messages: [system_role],
    temperature: temp,
    stream: true
  }

  if response_history[:choices].length > 1
    response_history[:choices][1..-1].each do |message|
      http_body[:messages].push(message)
    end
  end

  http_body[:messages].push(user_role)

  timeout = opts[:timeout]
   = opts[:spinner]

  response = ollama_rest_call(
    http_method: :post,
    rest_call: rest_call,
    http_body: http_body,
    timeout: timeout,
    spinner: 
  )

  json_resp = JSON.parse(response, symbolize_names: true)
  assistant_resp = json_resp[:choices].first[:message]
  json_resp[:choices] = http_body[:messages]
  json_resp[:choices].push(assistant_resp)

  speak_answer = true if opts[:speak_answer]

  if speak_answer
    answer = assistant_resp[:content]
    text_path = "/tmp/#{SecureRandom.hex}.pwn_voice"
    # answer = json_resp[:choices].last[:text]
    # answer = json_resp[:choices].last[:content] if gpt
    File.write(text_path, answer)
    PWN::Plugins::Voice.text_to_speech(text_path: text_path)
    File.unlink(text_path)
  end

  json_resp
rescue StandardError => e
  raise e
end

.chat_with_tools(opts = {}) ⇒ Object

Supported Method Parameters

response = PWN::AI::Ollama.chat_with_tools( messages: 'required - full OpenAI-format messages array (system/user/assistant/tool)', tools: 'optional - OpenAI tools array [function:{...}]', tool_choice: 'optional - "auto" | "none" | function:{name:..}', model: 'optional - overrides PWN::Env[:ollama][:model]', temp: 'optional - temperature (defaults to PWN::Env[:ollama][:temp] || 1)', timeout: 'optional - seconds (default 900)', spinner: 'optional - display spinner (default false)' )

Returns a Hash with :choices / :assistant_message intact (including :message) — used by PWN::AI::Agent::Loop.

LOCAL-MODEL SCAFFOLDING

This hits Ollama's NATIVE /api/chat (not the OpenAI-compat shim) so the following actually take effect:

options.num_ctx     - Ollama defaults to 2048; the pwn-ai system
                    prompt alone blows that. Defaults here to
                    PWN::Env[:ai][:ollama][:num_ctx] || 32768.
options.temperature - forced to 0.1 on tool-bearing turns for
                    deterministic tool selection; engine[:temp]
                    (creative) on the final text-only turn.
format              - ONLY set when engine[:format] is explicitly
                    configured. Never default to 'json' when
                    tools: are present — that fights native
                    tool_calls and kills mid-loop tool use.
keep_alive: '30m'   - avoids reload latency between iterations.

tool_calls come back with function.arguments as a Hash (not a JSON string), which PWN::AI::Agent::Dispatch.parse_args handles. Streaming is ON (stream: true): ollama_rest_call assembles NDJSON chunks back into a single response so the return shape is unchanged.



411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
 
# File 'lib/pwn/ai/ollama.rb', line 411

public_class_method def self.chat_with_tools(opts = {})
  engine   = PWN::Env[:ai][:ollama]
  messages = opts[:messages]
  raise 'ERROR: messages array is required' if messages.nil? || messages.empty?

  model = opts[:model] ||= engine[:model]
  raise 'ERROR: Model is required.  Call #get_models method for details' if model.nil?

  temp = opts[:temp].to_f
  temp = engine[:temp].to_f.nonzero? || 1 if temp.zero?

  tools_present = opts[:tools] && !opts[:tools].empty?
  tool_temp     = (engine[:tool_temp] || 0.1).to_f
  num_ctx       = (engine[:num_ctx]   || 32_768).to_i
  # Hard cap generation length. Thinking models (Qwen3 / R1-style)
  # otherwise stream unbounded message.thinking tokens until the
  # idle read_timeout (default 900s) — which looks like a 15–25 min
  # "stuck spinner" while bytes keep arriving on the socket.
  num_predict   = (engine[:num_predict] || 4_096).to_i
  keep_alive    = engine[:keep_alive] || '30m'

  http_body = {
    model: model,
    messages: messages,
    stream: true,
    keep_alive: keep_alive,
    options: {
      num_ctx: num_ctx,
      num_predict: num_predict,
      temperature: tools_present ? tool_temp : temp
    }
  }
  if tools_present
    http_body[:tools] = opts[:tools]
    # 0.2 — omit format when tools are present unless the operator
    # explicitly set PWN::Env[:ai][:ollama][:format]. Forcing 'json'
    # races the native tool_calls sampler and produces "can't call tools"
    # mid-loop on many local models.
    fmt = engine[:format]
    http_body[:format] = fmt unless fmt.nil? || fmt.to_s.empty?
  end
  http_body[:tool_choice] = opts[:tool_choice] if opts[:tool_choice]

  response = ollama_rest_call(
    http_method: :post,
    rest_call: 'ollama/api/chat',
    http_body: http_body,
    timeout: opts[:timeout],
    spinner: opts[:spinner]
  )
  raise 'ERROR: Ollama chat_with_tools received empty response from ollama_rest_call' if response.nil? || (response.respond_to?(:empty?) && response.empty?)

  json_resp = JSON.parse(response, symbolize_names: true)
  # Normalise native /api/chat shape to what Loop.normalize_llm expects.
  msg = json_resp[:message] || json_resp.dig(:choices, 0, :message)
  if msg.is_a?(Hash)
    content = msg[:content].to_s
    thinking = msg[:thinking].to_s
    tcalls = Array(msg[:tool_calls])
    msg = msg.merge(content: visible_from_thinking(thinking: thinking)) if content.strip.empty? && !thinking.strip.empty? && tcalls.empty?
  end
  json_resp[:choices] = [{ message: msg }] if msg && !json_resp.key?(:choices)
  json_resp[:assistant_message] = msg
  raise "ERROR: Ollama response missing message/choices: #{json_resp.inspect[0, 400]}" if msg.nil?

  json_resp
rescue StandardError => e
  raise e
end

.get_models ⇒ Object

Supported Method Parameters

response = PWN::AI::Ollama.get_models



369
370
371
372
373
374
375
 
# File 'lib/pwn/ai/ollama.rb', line 369

public_class_method def self.get_models
  models = ollama_rest_call(rest_call: 'ollama/api/tags')

  JSON.parse(models, symbolize_names: true)[:models]
rescue StandardError => e
  raise e
end

.help ⇒ Object

Display Usage for this Module



585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
 
# File 'lib/pwn/ai/ollama.rb', line 585

public_class_method def self.help
  puts "USAGE:
    models = #{self}.get_models

    response = #{self}.chat(
      request: 'required - message to Ollama',
      model: 'optional - model to use for text generation (defaults to PWN::Env[:ai][:ollama][:model])',
      temp: 'optional - creative response float (defaults to PWN::Env[:ai][:ollama][:temp])',
      system_role_content: 'optional - context to set up the model behavior for conversation (Default: PWN::Env[:ai][:ollama][:system_role_content])',
      response_history: 'optional - pass response back in to have a conversation',
      speak_answer: 'optional speak answer using PWN::Plugins::Voice.text_to_speech (Default: nil)',
      timeout: 'optional - timeout in seconds (defaults to 900)',
      spinner: 'optional - display spinner (defaults to false)'
    )

    #{self}.authors
  "
end