Posthook Metrics

AI agent sessions, prompts, files and attributed commits captured by posthook, as aggregates or row by row

What posthook data looks like

Posthook runs on each engineer’s machine and records what Claude Code, Cursor and Codex CLI wrote, what the engineer prompted, and which of those lines were committed. Bilanc models it into four tables, all reachable through the metrics endpoint:

Table	Grain	Count metric
posthook_sessions	One row per agent session	posthook-sessions-count
posthook_session_messages	One row per prompt or assistant reply	posthook-session-message-counts
posthook_session_files	One row per file touched in a session	posthook-session-file-counts
posthook_session_commits	One row per commit a session contributed to	posthook-session-commit-counts

Two more posthook metrics live on the pull_requests table because they measure AI lines that reached merged PRs: posthook-merged-ai-lines and posthook-ai-pct-of-code.

All four posthook tables carry the session owner’s Bilanc identity (merged_user_id, team_ids, squad and manager details), so every people filter and the same row-level security apply as elsewhere.

Only sessions from engineers whose posthook identity is mapped to a Bilanc user are returned. Sessions from unmapped emails are excluded, matching the dashboard.

Metrics

Session summary metrics (posthook_sessions)

posthook-sessions-count · count

Distinct sessions that generated or committed at least one AI line. Response column posthook_sessions_count. Matches the dashboard’s Active Sessions card. Sessions with no captured AI activity exist in the table but are not counted.

posthook-active-users · count

Distinct engineers with at least one session in the window. Response column posthook_active_users.

posthook-active-time · aggregation required

AI working time in seconds, derived from hook events: each session is active for every 5-minute window containing at least one event. Response column posthook_active_time. Pass "aggregation": "SUM"; the value is always a sum regardless.

posthook-total-tokens · aggregation required

Input + output tokens. Response column {agg}_total_tokens. Cache-read and cache-creation tokens are excluded, matching the posthook dashboard. The raw per-session total_tokens column includes them.

posthook-output-tokens · aggregation required

Output tokens only. Response column {agg}_output_tokens.

posthook-generated-lines · aggregation required

Lines the agent wrote, from edit events. Response column {agg}_generated_lines.

posthook-committed-lines · aggregation required

Lines the agent wrote that posthook attributed into commits. Response column {agg}_committed_lines.

posthook-ai-code-ratio · ratio

Committed ÷ generated lines. Response column posthook_ai_code_ratio.

posthook-top-model · string

Most-used model among sessions that committed AI code, ignoring placeholder slugs like default and unknown. Response column posthook_top_model.

Detail-table counts

posthook-session-message-counts · count

Distinct messages in posthook_session_messages. Response column posthook_session_message_counts.

posthook-session-file-counts · count

Rows in posthook_session_files. Response column posthook_session_file_counts.

posthook-session-commit-counts · count

Distinct commits in posthook_session_commits. Response column posthook_session_commit_counts.

PR-level attribution (pull_requests)

posthook-merged-ai-lines · aggregation required

AI-attributed lines that landed in merged PRs. Response column {agg}_merged_ai_lines. Uses PR date fields (pr_merged_at, pr_created_at, …) and PR filters.

posthook-ai-pct-of-code · ratio

AI committed lines ÷ lines added across the selected PRs, capped at 1. Response column posthook_ai_pct_of_code. The dashboard’s “AI % of Code” card requests this with "pr_states": ["merged"] and "date_field": "pr_merged_at".

Columns

Any column can be used in group_by and order_by; date-typed columns can be used in date_field.

posthook_sessions

Column	Type	Description
session_id	text	Unique session id
engineer_email	text	Git email posthook captured for the session
engineer_name	text	Engineer name posthook captured
agent_slug	text	Agent used: claude-code, cursor, codex, …
model_slug	text	Model used, as reported by the agent
model_provider	text	Vendor derived from model_slug: Anthropic, OpenAI, Google, xAI, DeepSeek, Alibaba, Other
repo_id	text	Posthook repository id
repo_name	text	Repository name
remote_url	text	Repository remote URL
branch	text	Branch the session worked on
started_at	timestamp	Session start
ended_at	timestamp	Session end (null for Cursor, which has no stop hook)
date	date	Date of started_at. Default date_field
input_tokens	integer	Input tokens
output_tokens	integer	Output tokens
cache_read_tokens	integer	Cache-read tokens
cache_creation_tokens	integer	Cache-creation tokens
total_tokens	integer	Sum of all four token columns
active_minutes	integer	Event-derived working time (distinct 5-minute windows × 5)
generated_lines	numeric	AI-written lines captured from edit events
files_edited	integer	Distinct files the agent edited
committed_lines	integer	AI lines attributed into commits
commits_contributed	integer	Commits the session contributed to
user_id	text	Posthook raw user id
merged_user_id	text	Resolved Bilanc user id
propelauth_user_id	text	Bilanc auth user id
name	text	Resolved Bilanc user name
role, location, level	text	From the Bilanc user profile
departments, team_names, team_ids	array	Org membership of the user

posthook_session_messages

Column	Type	Description
message_id	text	Prompt id for user rows; Stop event id for assistant rows
session_id	text	Owning session
message_seq	integer	1-based chronological position within the session
role	text	user or assistant
source	text	transcript or hook for user rows, stop_hook for assistant rows
agent_slug	text	Agent the message belongs to
message_text	text	Full prompt as typed, or the final assistant message of the turn
message_ts	timestamp	When the message was sent
tool_calls	integer	User rows only: tool-use events fired during the turn. Null for assistant rows and for agents without event capture (Cursor)
turn_minutes	integer	User rows only: minutes from the prompt to the turn’s final Stop event
date	date	Date of message_ts, falling back to the session date. Default date_field
engineer_email, merged_user_id, team_names, team_ids		Owner identity, copied from the session

posthook_session_files

Column	Type	Description
session_id	text	Owning session
file_path	text	Repo-relative path
edits	integer	AI edit events on this file in this session
lines_generated	numeric	AI lines generated on this file
lines_committed	integer	AI lines from this session attributed into commits touching this file
commits_touched	integer	Distinct commits this session contributed to on this file
first_edit_ts	timestamp	First AI edit (null when attribution came without a captured edit)
last_edit_ts	timestamp	Last AI edit
date	date	Date of first edit, falling back to last edit, then session date. Default date_field
engineer_email, merged_user_id, team_names, team_ids		Owner identity, copied from the session

posthook_session_commits

Column	Type	Description
session_id	text	Owning session
commit_id	text	Posthook commit id
sha	text	Git SHA
short_sha	text	First 7 characters of the SHA
branch	text	Branch the commit was made on
message	text	Commit message
committed_at	timestamp	Commit timestamp
author_email	text	Commit author email
lines_added	integer	Total lines the commit added
lines_removed	integer	Total lines the commit removed
files_changed	integer	Files the commit changed
lines_attributed	integer	AI lines posthook attributed from this session into this commit
attribution_source	text	How posthook matched the session to the commit
confidence	text	Attribution confidence
files_touched	integer	Files this session touched within the commit
pr_id	text	PR the commit landed in, when Bilanc has one (one PR kept per commit)
pr_source	text	VCS of the linked PR: github, gitlab, bitbucket, azure_devops
pr_title, pr_url, pr_state	text	Linked PR details
date	date	Date of committed_at, falling back to session date. Default date_field
engineer_email, merged_user_id, team_names, team_ids		Owner identity, copied from the session

Example: team summary by agent, week over week

This is the request behind the Posthook team table in the dashboard. Note posthook-ai-pct-of-code is a PR-table metric, so it gets pr_merged_at as its date field and the pr_states filter; the session-table metrics ignore pr_states.

Request

curl -X POST 'https://api.bilanc.co/metrics/get-multiple-metrics' \
  -H 'Authorization: YOUR_API_KEY' \
  -H 'Content-Type: application/json' \
  -d '{
    "filters": {
      "start_date": "2026-08-24",
      "end_date": "2026-09-06",
      "pr_states": ["merged"]
    },
    "group_by": ["agent_slug"],
    "date_level": "week",
    "metrics": [
      "posthook-sessions-count",
      "posthook-active-users",
      "posthook-total-tokens",
      "posthook-generated-lines",
      "posthook-committed-lines",
      "posthook-ai-code-ratio",
      "posthook-top-model",
      "posthook-ai-pct-of-code"
    ],
    "date_fields": {
      "posthook-sessions-count": "date",
      "posthook-active-users": "date",
      "posthook-total-tokens": "date",
      "posthook-generated-lines": "date",
      "posthook-committed-lines": "date",
      "posthook-ai-code-ratio": "date",
      "posthook-top-model": "date",
      "posthook-ai-pct-of-code": "pr_merged_at"
    },
    "aggregations": {
      "posthook-total-tokens": "SUM",
      "posthook-generated-lines": "SUM",
      "posthook-committed-lines": "SUM"
    },
    "include_previous_period": true
  }'

Cursor reports no token counts through its hooks, so token metrics for agent_slug = cursor are zero by design.

Example: every session as a row

Group by session_id and the columns you want. This is the request behind the dashboard’s session list. metric_min_value: 1 keeps only sessions that generated or committed AI code; drop it to include every captured session.

Request

curl -X POST 'https://api.bilanc.co/metrics/posthook-sessions-count' \
  -H 'Authorization: YOUR_API_KEY' \
  -H 'Content-Type: application/json' \
  -d '{
    "filters": {
      "start_date": "2026-09-01",
      "end_date": "2026-09-07"
    },
    "group_by": [
      "session_id", "name", "engineer_email", "agent_slug", "model_slug",
      "model_provider", "repo_name", "branch", "started_at", "ended_at",
      "active_minutes", "generated_lines", "committed_lines", "total_tokens",
      "input_tokens", "output_tokens", "files_edited", "commits_contributed"
    ],
    "date_field": "date",
    "metric_min_value": 1,
    "order_by": "started_at",
    "order_direction": "DESC",
    "limit": 200
  }'

For more than 200 sessions, window by day or use get-multiple-metrics. See Paging through large tables.

Example: a session’s prompts and replies

Pin the query to one session with filters.session_id. Set start_date a day or two before the session started: the date column of a message falls back to the session date, but commits attributed later can land after the session’s own day.

Request

curl -X POST 'https://api.bilanc.co/metrics/posthook-session-message-counts' \
  -H 'Authorization: YOUR_API_KEY' \
  -H 'Content-Type: application/json' \
  -d '{
    "filters": {
      "start_date": "2026-09-03",
      "session_id": "7f0c3b8e-2d1a-4b6f-9e21-0a5c6d4e8b90"
    },
    "group_by": ["message_id", "message_seq", "message_ts", "role", "source", "message_text", "tool_calls", "turn_minutes"],
    "date_field": "date",
    "order_by": "message_seq",
    "order_direction": "ASC"
  }'

To export prompts in bulk rather than per session, drop session_id and window by date instead, adding session_id and engineer_email to group_by so each message keeps its context.

Example: files touched and commits attributed

The same shape, against the other two detail tables. These are the dashboard’s “Files” and “Commits” panels.

Files

curl -X POST 'https://api.bilanc.co/metrics/posthook-session-file-counts' \
  -H 'Authorization: YOUR_API_KEY' \
  -H 'Content-Type: application/json' \
  -d '{
    "filters": {
      "start_date": "2026-09-03",
      "session_id": "7f0c3b8e-2d1a-4b6f-9e21-0a5c6d4e8b90"
    },
    "group_by": ["file_path", "edits", "lines_generated", "lines_committed", "commits_touched", "first_edit_ts", "last_edit_ts"],
    "date_field": "date",
    "order_by": "lines_generated",
    "order_direction": "DESC"
  }'

Example: daily usage per engineer

A conventional aggregate: tokens, active time and committed lines per engineer per day.

curl -X POST 'https://api.bilanc.co/metrics/get-multiple-metrics' \
  -H 'Authorization: YOUR_API_KEY' \
  -H 'Content-Type: application/json' \
  -d '{
    "filters": {"start_date": "2026-09-01", "end_date": "2026-09-30"},
    "group_by": ["name", "engineer_email"],
    "date_level": "day",
    "metrics": ["posthook-sessions-count", "posthook-active-time", "posthook-total-tokens", "posthook-committed-lines"],
    "date_fields": {
      "posthook-sessions-count": "date",
      "posthook-active-time": "date",
      "posthook-total-tokens": "date",
      "posthook-committed-lines": "date"
    },
    "aggregations": {
      "posthook-active-time": "SUM",
      "posthook-total-tokens": "SUM",
      "posthook-committed-lines": "SUM"
    }
  }'

Parameter notes

date_field · string

All four posthook tables default to date. Sessions also accept started_at and ended_at; messages message_ts; files first_edit_ts and last_edit_ts; commits committed_at. The two PR-table metrics take PR date fields (pr_merged_at, pr_created_at, pr_closed_at, …).

filters · object

Posthook tables honour member, team, departments, locations, levels, squad_ids, squad_levels, manager_ids and session_id. They have no repository or PR-state columns, so repositories and pr_states are ignored for them (filter on the repo_name column via group_by instead). posthook-merged-ai-lines and posthook-ai-pct-of-code are PR-table metrics and honour every PR filter.

aggregation · string

Required for posthook-total-tokens, posthook-output-tokens, posthook-generated-lines, posthook-committed-lines, posthook-active-time and posthook-merged-ai-lines. Use SUM for totals and AVG for per-session averages. Count, ratio and top-model metrics take none.