RawMessageRecord#

Restore model context and preserve immutable original messages. See the guide.

class zett_agent.extensions.persistence.RawMessageRecord(id, session_id, request_id, sequence, message, metadata, tags, started_at, completed_at, duration_ns, reasoning_started_at, reasoning_completed_at, reasoning_duration_ns, content_started_at, content_completed_at, content_duration_ns, input_tokens, output_tokens, cache_read_tokens, cache_write_tokens, reasoning_tokens, created_at, updated_at)[source]#

One immutable original message returned by session storage.

Parameters:

  • id (str)

  • session_id (str)

  • request_id (str)

  • sequence (int)

  • message (AnyMessage)

  • metadata (dict[str, JsonValue])

  • tags (dict[str, JsonValue])

  • started_at (datetime)

  • completed_at (datetime)

  • duration_ns (int)

  • reasoning_started_at (datetime | None)

  • reasoning_completed_at (datetime | None)

  • reasoning_duration_ns (int | None)

  • content_started_at (datetime | None)

  • content_completed_at (datetime | None)

  • content_duration_ns (int | None)

  • input_tokens (int | None)

  • output_tokens (int | None)

  • cache_read_tokens (int | None)

  • cache_write_tokens (int | None)

  • reasoning_tokens (int | None)

  • created_at (datetime)

  • updated_at (datetime)

metadata: dict[str, JsonValue]#

Request-scoped application data captured independently of model messages.

tags: dict[str, JsonValue]#

Request-scoped classifications captured independently of model messages.

started_at: datetime#

UTC operation start; GENERATING for assistants, RUNNING_TOOL for tools, and append time for user-authored or directly imported messages.

completed_at: datetime#

UTC operation completion; equal to started_at for instantaneous messages.

duration_ns: int#

Total operation duration in nanoseconds, measured by a monotonic clock.

reasoning_started_at: datetime | None#

UTC arrival time of the first streamed reasoning delta, when present.

reasoning_completed_at: datetime | None#

UTC boundary where reasoning ended before content, tools, or completion.

reasoning_duration_ns: int | None#

Monotonic elapsed reasoning time in nanoseconds; None if not observed.

content_started_at: datetime | None#

UTC arrival time of the first streamed answer-content delta, when present.

content_completed_at: datetime | None#

UTC final-response boundary for a streamed answer-content segment.

content_duration_ns: int | None#

Monotonic elapsed content-streaming time in nanoseconds; None if absent.

input_tokens: int | None#

Provider-reported prompt tokens for an assistant model response.

output_tokens: int | None#

Provider-reported completion tokens, including reasoning when applicable.

cache_read_tokens: int | None#

Input tokens served from a provider cache.

cache_write_tokens: int | None#

Input tokens written into a provider cache when separately reported.

reasoning_tokens: int | None#

Reasoning tokens included within output_tokens when separately reported.

property total_tokens: int | None#

Return input plus output tokens when provider usage was recorded.

property cache_hit_rate: float | None#

Return a zero-to-one input cache ratio, or None without input usage.