turnstone/docs/diagrams/03-core-engine-classes.puml

@startuml
!theme plain
title Turnstone — Core Engine Classes

skinparam classAttributeIconSize 0

' SessionUI Protocol
interface "SessionUI" as SessionUI <<Protocol>> {
    + on_thinking_start()
    + on_thinking_stop()
    + on_reasoning_token(text: str)
    + on_content_token(text: str)
    + on_stream_end()
    + approve_tools(items: list) → (bool, str|None)
    + on_tool_result(call_id: str, name: str, output: str)
    + on_tool_output_chunk(call_id: str, chunk: str)
    + on_status(usage: dict, ctx_window: int, effort: str)
    + on_plan_review(content: str) → str
    + on_info(message: str)
    + on_error(message: str)
    + on_state_change(state: str)
    + on_rename(name: str)
}

' Implementations
class "TerminalUI" as TerminalUI {
    Writes to stdout with ANSI colors
    Prompts for approval via input()
    --
    cli.py
}

class "WorkstreamTerminalUI" as WsTermUI {
    - _output_buffer: list[tuple]
    - ws_id: str
    - manager: WorkstreamManager
    + flush_buffer()
    --
    Buffers output when workstream
    is not foregrounded
}

class "WebUI" as WebUI {
    - _event_queue: Queue
    - _approval_event: Event
    - _plan_event: Event
    - _ws_prompt_tokens: int
    - _ws_tool_calls: dict
    + resolve_approval(approved, feedback)
    + resolve_plan(feedback)
    --
    Enqueues JSON events for SSE.
    Blocks on threading.Event for
    approval/plan review.
    SSE handlers bridge Queue to
    async via run_in_executor().
    --
    server.py
}

class "NullUI" as NullUI {
    approve_tools() → (True, None)
    All other methods: no-op
    --
    eval.py
}

' LLMProvider Protocol
interface "LLMProvider" as LLMProvider <<Protocol>> {
    + provider_name: str {property}
    + get_capabilities(model) → ModelCapabilities
    + create_streaming(client, model, messages, ...) → Iterator[StreamChunk]
    + create_completion(client, model, messages, ...) → CompletionResult
    + convert_tools(tools) → list[dict]
    + retryable_error_names: frozenset[str] {property}
    --
    core/providers/_protocol.py
}

class "OpenAIProvider" as OpenAIProv {
    Model capability lookup table
    (GPT-5.x, O-series, search)
    Passthrough: messages already
    in OpenAI format.
    Search models: web_search_options
    + url_citation annotations.
    --
    core/providers/_openai.py
}

class "AnthropicProvider" as AnthropicProv {
    Converts OpenAI messages to
    Anthropic content blocks.
    Adaptive + manual thinking.
    Native web search via
    web_search_20250305 server tool.
    Lazy anthropic SDK import.
    --
    core/providers/_anthropic.py
}

' ModelCapabilities
class "ModelCapabilities" as ModelCaps <<frozen>> {
    + context_window: int
    + max_output_tokens: int
    + supports_temperature: bool
    + token_param: str
    + thinking_mode: str
    + supports_effort: bool
    + supports_web_search: bool
}

' ChatSession
class "ChatSession" as ChatSession {
    - client: Any
    - provider: LLMProvider
    - model: str
    - ui: SessionUI
    - messages: list[dict]
    - _msg_tokens: list[int]
    - _session_id: str
    - _mcp_client: MCPClientManager | None
    - _registry: ModelRegistry | None
    + model_alias: str | None {property}
    - _tools: list[dict]
    - _task_tools: list[dict]
    - _agent_tools: list[dict]
    - _read_files: set[str]
    - system_messages: list[dict]
    --
    + send(user_input: str)
    + handle_command(command: str)
    + resume_session(session_id: str)
    - _save_config()
    - _stream_response(stream) → dict
    - _create_stream_with_retry(msgs) → Stream (+ fallback)
    - _try_stream(client, model, msgs) → Stream
    - _execute_tools(tool_calls) → (results, feedback)
    - _prepare_tool(tc) → item dict
    - _prepare_mcp_tool(call_id, name, args) → item dict
    - _exec_mcp_tool(item) → (call_id, output)
    - _run_agent(messages, tools, ...) → str
    - _compact_messages(auto: bool)
    - _full_messages() → list[dict]
    - _update_token_table(msg)
    - _emit_state(state: str)
    - _generate_title()
}

' HeadlessSession
class "HeadlessSession" as HeadlessSession {
    + tool_call_log: list[dict]
    + auto_approve: bool = True
    + send_headless(input, max_turns, ...)
    - _override_system_prompt(content)
    --
    eval.py: non-streaming,
    records all tool calls
}

' WorkstreamManager
class "WorkstreamManager" as WsMgr {
    - _session_factory: Callable[[SessionUI], ChatSession]
    - _workstreams: dict[str, Workstream]
    - _order: list[str]
    - _active_id: str
    - _on_state_change: Callable
    --
    + create(name, ui_factory) → Workstream
    + close(ws_id)
    + get(ws_id) → Workstream
    + get_active() → Workstream
    + switch(ws_id)
    + set_state(ws_id, state)
    + close_idle(max_age_seconds)
    - _max_workstreams: int
}

' Workstream
class "Workstream" as Ws <<dataclass>> {
    + id: str
    + name: str
    + state: WorkstreamState
    + session: ChatSession
    + ui: SessionUI
    + worker_thread: Thread
    + error_message: str
    + last_active: float
    - _lock: Lock
}

' WorkstreamState
enum "WorkstreamState" as WsState {
    IDLE
    THINKING
    RUNNING
    ATTENTION
    ERROR
}

' MCPClientManager
class "MCPClientManager" as MCPMgr {
    - _sessions: dict[str, ClientSession]
    - _tools: list[dict]
    - _tool_map: dict[str, tuple]
    --
    + start()
    + get_tools() → list[dict]
    + is_mcp_tool(name) → bool
    + call_tool_sync(name, args) → str
    + shutdown()
    --
    Background asyncio event loop
    bridges async MCP SDK to
    sync ChatSession dispatch.
    --
    core/mcp_client.py
}

' ModelRegistry
class "ModelRegistry" as ModelReg {
    - _models: dict[str, ModelConfig]
    - _clients: dict[str, Any]
    - _providers: dict[str, LLMProvider]
    - _client_lock: Lock
    + default: str
    + fallback: list[str]
    + agent_model: str | None
    --
    + resolve(alias) → (client, model, config)
    + get_client(alias) → Any
    + get_provider(alias) → LLMProvider
    + has_alias(alias) → bool
    + list_aliases() → list[str]
    + shutdown()
    --
    Thread-safe lazy client + provider
    creation. Loaded by load_model_registry()
    from CLI args + [models.*] config.
    --
    core/model_registry.py
}

class "ModelConfig" as ModelCfg <<frozen>> {
    + alias: str
    + provider: str
    + base_url: str
    + model: str
    + context_window: int
}

' Circuit breaker state
enum "CircuitState" as CircuitState {
    CLOSED
    OPEN
    HALF_OPEN
}

' BackendHealthMonitor
class "BackendHealthMonitor" as HealthMon {
    - _state: CircuitState
    - _consecutive_failures: int
    - _probe_interval: float
    - _probe_timeout: float
    - _threshold: int
    - _cooldown: float
    --
    + start()
    + stop()
    + record_success()
    + record_failure()
    + should_allow_request() → bool
    --
    Daemon thread probes
    client.models.list() on interval.
    --
    core/healthcheck.py
}

' RateLimiter
class "RateLimiter" as RateLimiter {
    - _buckets: dict[str, TokenBucket]
    - _rate: float
    - _burst: int
    --
    + check(client_ip: str, path: str) → (bool, float)
    --
    Per-IP token bucket.
    Returns (allowed, retry_after).
    --
    core/ratelimit.py
}

' TokenBucket
class "TokenBucket" as TokenBucket {
    - _tokens: float
    - _rate: float
    - _burst: int
    - _last_refill: float
    --
    + consume() → bool
    + retry_after: float {property}
    --
    Single-IP bucket with
    refill rate and burst cap.
}

' Relationships
SessionUI <|.. TerminalUI
TerminalUI <|-- WsTermUI
SessionUI <|.. WebUI
SessionUI <|.. NullUI

LLMProvider <|.. OpenAIProv
LLMProvider <|.. AnthropicProv

ChatSession --> SessionUI : uses
ChatSession --> LLMProvider : delegates LLM calls
ChatSession --> MCPMgr : optional
ChatSession --> ModelReg : optional
ChatSession <|-- HeadlessSession

WsMgr --> "*" Ws : manages
Ws --> "1" ChatSession : wraps
Ws --> "1" SessionUI : wraps
Ws --> "1" WsState : has

WsMgr ..> ChatSession : creates via\nsession_factory(ui, model_alias)

ModelReg --> "*" ModelCfg : holds
ModelReg --> "*" LLMProvider : caches
LLMProvider --> ModelCaps : returns

ChatSession --> HealthMon : checks circuit
HealthMon --> "1" CircuitState : has
RateLimiter --> "*" TokenBucket : per IP

note bottom of ChatSession
  Central engine: multi-turn LLM loop
  with tool dispatch, agent sub-sessions,
  context compaction, and memory persistence.
  Provider-agnostic — delegates all LLM
  communication to LLMProvider adapters.

  core/session.py (~2700 lines)
end note

@enduml