package v1.public.events.copilot
Get desktop application:
View/edit binary Protocol Buffers messages
message
copilot.proto:197Fired when context compaction runs within an agent run. Emitted for every compaction, including tool_result_clearing which makes no LLM call, so it captures compaction frequency and effectiveness the summarizer's CopilotLlmCall alone cannot. Each strategy fires at most once per run, so copilot_request_id is the key: it identifies the run and joins a "summarization" event to the priced summarizer CopilotLlmCall (the single row tagged call_purpose="compaction" in that run).
optional enrichment = 1
Server enrichment for this event
optional ctx = 2
dbt Cloud context (account_id, user_id, project_id, etc.)
string copilot_request_id = 3
Identifies the parent agent run; foreign key to CopilotGenerationComplete. Also the cost join key: a "summarization" event maps 1:1 to the run's single CopilotLlmCall tagged call_purpose="compaction".
string thread_id = 4
Conversation thread for multi-turn correlation
string strategy = 5
Which compaction strategy ran; see strategy values documented above
string trigger = 6
What triggered this compaction; see trigger values documented above
optional int64 tokens_before = 7
Approximate context size before compaction, in the middleware's working token estimate (count_tokens_approximately), the same space as compaction_threshold_used. Not provider-billed tokens; the summarizer's real cost is on its CopilotLlmCall.
optional int64 tokens_after = 8
Approximate context size after compaction, same estimate space as tokens_before. tokens_before minus tokens_after is the context reclaimed.
optional int64 compaction_threshold_used = 9
Working token limit that triggered automatic compaction, same estimate space as tokens_before/after. Per model (derived from the context window; conservative fallback for opaque/BYOK models). Unset when user_forced.
optional common_context = 10
message
copilot.proto:23Fired when an AI code generation request completes (success or failure). Covers all Copilot actions: generate docs, tests, semantic models, metrics, inline SQL, text-to-SQL, and agent runs.
optional enrichment = 1
Server enrichment for this event
optional ctx = 2
dbt Cloud context (account_id, user_id, project_id, etc.)
string copilot_request_id = 3
Unique identifier for this generation request; natural dedup key
string thread_id = 4
Conversation thread for multi-turn correlation
string model_provider = 5
LLM provider (e.g. "openai", "azure_openai", "anthropic")
string model_version = 6
Model identifier (e.g. "gpt-4o", "claude-sonnet-4-20250514")
string app_name = 7
Client application name (e.g. "studio", "vscode", "cloud-cli")
string copilot_trigger = 8
What triggered the generation (e.g. "user", "auto")
string copilot_action = 9
Which Copilot action ran (e.g. "generate_documentation", "text_to_sql", "agent")
bool has_error = 10
Whether the generation encountered an error
optional string error_message = 11
Error message if has_error is true
bool has_custom_styleguide = 12
Whether the project has a custom style guide configured
string prompt_version = 13
Prompt template version hash
string prompt_name = 14
Prompt template name
string prompt_variation = 15
Prompt template variation (e.g. A/B test variant)
optional int64 tokens_in_prompt = 16
Input tokens sent to the LLM
optional int64 tokens_in_response = 17
Output tokens received from the LLM
optional int64 tokens_total = 18
Total tokens (input + output)
optional int64 latency_ms = 19
End-to-end latency in milliseconds
optional string ide_session_id = 20
IDE/develop session identifier
string key_type = 21
API key type; see key_type values documented above
optional int64 cache_creation_input_tokens = 22
Input tokens where prompt cache was created (cache miss, Anthropic)
optional int64 cache_read_input_tokens = 23
Input tokens served from prompt cache (cache hit)
optional int64 reasoning_output_tokens = 24
Chain-of-thought tokens not included in the response
optional int64 first_token_ms = 25
Time to first token in milliseconds. Unset when not streaming.
optional common_context = 26
message
copilot.proto:109Fired once per LLM call within an agent loop. Allows per-call token and latency attribution when a single copilot_request_id triggers multiple LLM invocations (e.g. a dev agent loop). Links back to the parent CopilotGenerationComplete via copilot_request_id.
optional enrichment = 1
Server enrichment for this event
optional ctx = 2
dbt Cloud context (account_id, user_id, project_id, etc.)
string copilot_request_id = 3
Identifies the parent agent run; foreign key to CopilotGenerationComplete
string thread_id = 4
Conversation thread for multi-turn correlation
string model_provider = 5
LLM provider (e.g. "openai", "azure_openai", "anthropic")
string model_version = 6
Model identifier (e.g. "gpt-4o", "claude-sonnet-4-20250514")
string app_name = 7
Client application name (e.g. "studio", "vscode", "cloud-cli")
string copilot_action = 8
Which Copilot action this call belongs to (e.g. "run_langgraph_agent")
string key_type = 9
API key type used for this call. String representation of the DbtCopilotGenerationComplete.KeyType enum (private proto); valid values are the same as CopilotGenerationComplete.key_type documented above. Enums are not yet handled properly in Vortex, so the enum name is serialised to a string at the call site rather than using an enum field.
optional int32 call_index = 10
Zero-based index of this LLM call within the agent loop
optional int64 tokens_input = 11
Input tokens sent to the LLM for this call
optional int64 tokens_output = 12
Output tokens received from the LLM for this call
optional int64 tokens_total = 13
Total tokens (input + output)
optional int64 latency_ms = 14
Latency of this specific LLM call in milliseconds
optional int64 cache_creation_input_tokens = 15
Input tokens where prompt cache was created (cache miss, Anthropic)
optional int64 cache_read_input_tokens = 16
Input tokens served from prompt cache (cache hit)
optional int64 reasoning_output_tokens = 17
Chain-of-thought tokens not included in the response
optional int64 first_token_ms = 18
Time to first token in milliseconds. Unset when not streaming.
optional string call_purpose = 19
Purpose of this call within the parent run: "generation" - user-facing agent/generation turn (default when unset) "compaction" - context-compaction summarization call Unset is treated as "generation", so calls emitted before this field existed count as generation turns. Lets downstream models keep compaction out of turn counts while still pricing its token cost.
optional common_context = 20
message
copilot.proto:245Fired when a user responds to a Copilot suggestion.
optional enrichment = 1
Server enrichment for this event
optional ctx = 2
dbt Cloud context
string copilot_request_id = 3
The generation request this acceptance responds to
string thread_id = 4
Conversation thread for multi-turn correlation
string response_type = 5
User's decision; see response_type values documented above
optional string feedback_text = 6
Optional free-text feedback comment from the user. Unset when no feedback provided.
optional common_context = 7