Class: Mistri::Providers::OpenAI

Inherits:
Object
  • Object
show all
Defined in:
lib/mistri/providers/openai.rb,
lib/mistri/providers/openai/assembler.rb,
lib/mistri/providers/openai/serializer.rb

Overview

The OpenAI Responses API, streamed and stateless: store is always false, the full history replays every turn, and encrypted reasoning items round trip through the signature slots so nothing depends on server-side state. Reasoning summaries stream as thinking. max_output_tokens is deliberately omitted: the API defaults to the model's own ceiling.

Provider failures fold into the stream as an error turn rather than raising, matching the Anthropic provider's contract.

Defined Under Namespace

Modules: Serializer Classes: Assembler

Constant Summary collapse

DEFAULT_REASONING = 
{ summary: "auto" }.freeze

Instance Attribute Summary collapse

Instance Method Summary collapse

Constructor Details

#initialize(api_key:, model: "gpt-5.5", origin: "https://api.openai.com", reasoning: DEFAULT_REASONING, **transport_options) ⇒ OpenAI

Returns a new instance of OpenAI.



16
17
18
19
20
21
22
 
# File 'lib/mistri/providers/openai.rb', line 16

def initialize(api_key:, model: "gpt-5.5", origin: "https://api.openai.com",
               reasoning: DEFAULT_REASONING, **transport_options)
  @api_key = api_key
  @model = model
  @reasoning = reasoning
  @transport = Transport.new(origin: origin, **transport_options)
end

Instance Attribute Details

#model ⇒ Object (readonly)

Returns the value of attribute model.



24
25
26
 
# File 'lib/mistri/providers/openai.rb', line 24

def model
  @model
end

Instance Method Details

#close ⇒ Object 



41
 
# File 'lib/mistri/providers/openai.rb', line 41

def close = @transport.close

#stream(messages:, system: nil, tools: [], signal: nil, **overrides, &emit) ⇒ Object 



26
27
28
29
30
31
32
33
34
35
36
37
38
39
 
# File 'lib/mistri/providers/openai.rb', line 26

def stream(messages:, system: nil, tools: [], signal: nil, **overrides, &emit)
  model = overrides.fetch(:model, @model)
  assembler = OpenAI::Assembler.new(model: model)
  body = build_body(model, messages, system, tools, overrides)
  outcome = @transport.stream_post("/v1/responses", body: body, headers: headers,
                                                    signal: signal) do |record|
    assembler.feed(
      record, &emit
    )
  end
  outcome == :aborted ? assembler.abort(&emit) : assembler.finish(&emit)
rescue Error => e
  assembler.fail_stream(e, &emit)
end