Struct DatalogWriter 

pub struct DatalogWriter { /* private fields */ }

Accumulates log entries for a single turn, flushing to disk after each operation.

Implementations

impl DatalogWriter

pub fn new(working_dir: &Path, config: &DatalogConfig) -> Self

pub fn new_with_filename_tag( working_dir: &Path, config: &DatalogConfig, filename_tag: &str, ) -> Self

pub fn set_working_dir(&mut self, dir: &Path)

Update the working directory (e.g. after /cd). Absolute / ~/ configured paths are unaffected — only the default and relative cases follow cwd changes.

pub fn resolve_log_dir(base_dir: &Path, configured: Option<&str>) -> PathBuf

Resolve the actual directory to write datalog files into, given the current base_dir and configured_dir. Pure function — used by begin_turn, by runner.rs to decide where log_llm_request writes the JSONL request dump (so both writers stay in sync), and covered by unit tests.

The result is always <root>/<project-slug> so multiple projects don’t collide. <root> is $ATOMCODE_HOME/datalog by default, or whatever the user configured (absolute / ~/… / relative-to-cwd).

pub fn clear(&mut self)

Clear the current turn log state and delete the log file if it exists. Called when user runs /clear command.

pub fn begin_turn( &mut self, user_message: &str, model_name: &str, context_window: usize, )

Start a new turn: create log file, write env info + user message.

pub fn log_llm_call(&mut self)

Log start of a new LLM round-trip (increments the turn counter). Records the duration of the previous LLM turn.

pub fn log_context_stats( &mut self, system_tokens: usize, sent_tokens: usize, dropped_tokens: usize, _working_set_tokens: usize, total_messages: usize, )

Log context statistics for debugging.

pub fn log_cache_hit(&mut self, prompt_tokens: usize, cached_tokens: usize)

Log prompt cache hit info (only when provider reports cached_tokens > 0).

pub fn log_token_usage( &mut self, prompt_tokens: usize, completion_tokens: usize, cached_tokens: usize, )

Log token usage for the current LLM round-trip.

pub fn log_llm_dump( &mut self, messages: &[Message], tool_count: usize, model: &str, context_window: usize, )

Dump full LLM request as JSON into the datalog for debugging. Appends to a single JSONL file (one JSON object per line) colocated with the turn .md file: <turn_timestamp>_requests.jsonl. Each line has the step number so you can correlate with the md.

pub fn log_tool_call(&mut self, name: &str, args: &str)

Log a tool call start (within the current LLM turn).

pub fn log_tool_result(&mut self, output: &str, success: bool)

Log a tool call result.

pub fn log_model_text(&mut self, text: &str)

Log model text output between tool calls (plan, thinking, explanation).

pub fn log_text(&mut self, text: &str)

Log final assistant text output (response/summary).

pub fn log_error(&mut self, error: &str)

Log an error.

pub fn log_warning(&mut self, warning: &str)

Log a non-fatal advisory (e.g. provider truncation detector). Persisting it here makes post-hoc datalog inspection useful even when the live TUI line scrolls past — the user can grep the markdown for **Warning:** after the run.

pub fn end_turn(&mut self, total_tokens: usize, tool_call_count: usize)

End the turn: write duration and final flush.