Documentation
¶
Overview ¶
Package kimi drives the kimi CLI as a subprocess and parses its stream-json output into a plain text reply.
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
func ParseFinalMessage ¶
Kimi CLI persists token usage in internal session files (`wire.jsonl` StatusUpdate.payload.token_usage and `context.jsonl` `_usage` rows), but `kimi --print --output-format stream-json` does not emit those events on stdout. The installed JsonPrinter only forwards assistant messages, notifications, tool results, and plan displays, so agentbridge cannot obtain per-run token usage from the JSONL stream it parses here.
Types ¶
type Runner ¶
type Runner struct {
Command string
Timeout time.Duration
Env map[string]string
WorkspaceDir string
}
Runner executes the kimi CLI for a single request.
func (Runner) Run ¶
Run is a convenience wrapper that runs without thread resumption or progress callbacks.
func (Runner) RunWithThreadAndProgress ¶
func (r Runner) RunWithThreadAndProgress( ctx context.Context, threadID string, userText string, model string, env map[string]string, onProgress func(step string), onRawEvent func(kind, line, detail string), ) (string, string, error)
RunWithThreadAndProgress runs the kimi CLI and returns the final reply and next session ID.
- threadID: resume an existing session when non-empty.
- userText: the fully assembled prompt.
- model: overrides the CLI default when non-empty.
- env: merged over the process environment.
- onProgress: called with intermediate assistant messages; may be nil.
- onRawEvent: optional callback for raw stdout events (kind, line, detail); nil disables raw event delivery.
Source Files