Skip to content
Start free
Product Demo Pricing Security Docs Changelog Start free

Is AI making your team faster,
or just busier?

DevGlimpse shows, per team, whether your AI coding tools are shortening delivery or just producing bigger pull requests that wait longer for review. From read-only Git metadata. Every formula is public. Free to start. First numbers arrive within minutes of connecting; history fills in behind them.

Start free with GitHub

A GitHub organization admin installs the app. Not an admin? You can send them an invite.

Try the demo Book 20 minutes
  • Read-only access
  • Never stores your code
  • No individual rankings
Show which ink

This week 32 no AI signal 43 AI-assisted merged pull requests

A comparison, not a cause. “No AI signal” is not “no AI”: inline completions leave no trace.

Sample chart of merged pull requests per week over twelve weeks. Pull requests with no AI signal go from 46 to 32 a week while AI-assisted pull requests grow from 7 to 43.
Merged pull requests per week, AI-assisted vs no AI signal (sample data)
WeekNo AI signalAI-assisted
11 weeks ago 46 7
10 weeks ago 44 9
9 weeks ago 47 10
8 weeks ago 42 14
7 weeks ago 44 16
6 weeks ago 40 21
5 weeks ago 41 24
4 weeks ago 37 29
3 weeks ago 38 31
2 weeks ago 34 36
Last week 35 39
This week 32 43

Connect what you already use.

19 connections are live today. One model sits behind all of them, so cycle time means the same thing whether the work happened on GitHub, GitLab or Bitbucket.

Code

Pull requests, reviews, commits and teams.

  • GitHub
  • GitLab
  • Bitbucket Cloud

Tickets

Linked to the pull requests that close them.

  • Linear
  • Jira Cloud
  • Asana

Incidents and CI

Deploys, failures and time to restore from real events.

  • PagerDuty
  • incident.io
  • Opsgenie
  • GitHub Actions
  • GitLab pipelines

AI tools

Usage reported over OpenTelemetry, with a privacy allowlist.

  • Claude Code
  • Codex CLI
  • Gemini CLI
  • Org-level Copilot usage Building
  • Cursor Building
  • Anthropic and OpenAI usage Building

Sends to

Alerts, digests and answers, where your team already is.

  • Slack
  • Microsoft Teams
  • Email digests
  • MCP server
  • REST API By plan

Live Building now Everything on the roadmap

Minutes from install to insight.

No agents to deploy, no ticket hygiene to fix first, no six-week onboarding. Connect a code host and read your first dashboard the same day.

  1. Install the app

    An org admin installs DevGlimpse. We verify the install with the code host and create your workspace.

    • Metadata read
    • Pull requests read
    • Contents commit list only, never files or diffs read
    • Organization Members read
    • Deployments read
    • Actions read
    • Copilot metrics read

  2. The full 90 days, within hours

    DevGlimpse walks your history inside the host's rate limits, checkpoints every page, and shows an estimate.

    acme/payments-api 142 PRs
    acme/webqueued
    acme/mobilequeued

    Sample data

  3. A dashboard for every team

    Teams arrive from your code host on their own. Custom groups and an org hierarchy cover everything else.

    • Platform
    • Payments
    • Mobile
    • Data
    • Growth
    • SRE
    • + New group

    Sample data

Self-healing. Webhooks keep it live. An hourly redelivery and a nightly reconcile repair anything the host failed to deliver.

AI-assisted work, next to the rest.

Your teams may run more than one AI tool, and each vendor's dashboard only sees itself. DevGlimpse sees what actually merged, AI-assisted or with no AI signal, across Claude Code, Codex CLI and Gemini CLI. Org-level Copilot usage, Cursor and spend adapters are still being built.

AI-assisted against no-AI-signal pull requests

No AI signalAI-assisted
Measure No AI signal AI-assisted
Cycle time median 26h 19h
PR size lines changed 180 262
Review depth comments per PR 3.1 2.2
Revert rate of merged PRs 1.8% 2.7%
Linked bugs per 100 PRs 4.0 5.2

Sample data

From agent session to shipped change

  1. Agent session Claude Code, 41 min, 23 tool calls
  2. Pull request opened payments-api, 212 lines
  3. Human review 2 reviewers, 5 comments
  4. Merged and deployed linked with high confidence

Sample data

Adoption and spend
Who uses which assistant, how often, and what it costs, by team, tool, ticket and initiative.
Output against merged code
What the tools report writing next to what actually merged, with edit acceptance from their own telemetry.
Quality, compared
Reverts, linked bugs, incidents and review depth for AI-assisted work next to work with no AI signal.
Return on investment
Mark the day a team adopted a tool, then read delivery before and after against the spend.

A comparison, not a cause: “no AI signal” is not “no AI”, because inline completions leave no trace. AI numbers are shown per team and per tool, never as a score on a person. Each developer gets a private view of their own AI usage, and the telemetry receiver keeps only an allowlist of fields. Anything it does not recognise is dropped.

A report for every question a delivery review asks.

Deployments, incidents and tickets are linked to the pull requests behind them, so the answers come from real events, not surveys or guesses. Pick a question.

PR Workflow: time in each stage

  1. First commit to open 6.4h
  2. Open to first review 11h
  3. First review to merge 7.8h
  4. Merge to deploy 2.6h

Average time in stage The wait for a first review is the long pole here. Cycle time over the same window: median 21h, 90th percentile 2d 14h.

The four DORA keys, against industry bands

  • Deployment frequency 3.2 a day Elite
  • Lead time for changes 19h High
  • Change failure rate 6% High
  • Time to restore 48m Elite

Bands run from low on the left to elite on the right. Failure rate and time to restore come from your deploys, or from PagerDuty, incident.io and Opsgenie incidents.

CI Analytics: duration and failures by workflow

  • build 6.2m 2% fail
  • unit-tests 14.7m 9% fail
  • e2e 22.1m 14% fail
  • deploy 3.8m 1% fail

End-to-end tests take the longest and fail one run in seven. That is where a retry habit starts.

Predictability: done against committed

  • Cycle 31 35 of 42
  • Cycle 32 38 of 40
  • Cycle 33 33 of 46
  • Cycle 34 36 of 38
  • Cycle 35 39 of 41

Done Committed Rollover and scope creep sit on the same report, from Linear, Jira or Asana cycles.

Review load by team

  • Platform 118 / 74
  • Payments 64 / 92
  • Mobile 81 / 70
  • Data 37 / 58

Reviews given PRs opened Platform reviews far more than it opens. A team view, in place of an individual leaderboard.

Report builder

Metric
Review wait, median
Group by
Team
Filter
AI-assisted PRs, last 30 days
  • Platform 21h
  • Payments 34h
  • Mobile 27h
  • Data 16h

Compose a metric, a grouping and filters, then save it as a custom report, pin it to a dashboard or share a read-only link.

Sample data

Also live: Bug Creation, Bug Resolution and Bug Workload, Initiatives with a forecast, a collaboration network, work in progress, pulse surveys and custom dashboards. See every report

Team numbers for everyone. Your numbers for you.

Each role's view of individual metrics is a workspace setting, enforced on the server on every query. Rows outside your scope are never sent to your browser. Try it.

View as

Individual scope: Only me

You see your own numbers and every team aggregate. Nobody is ranked against a colleague.

Individual scope: My groups

Managers see the people in the teams and groups they belong to, plus every team aggregate.

Individual scope: Everyone

Admins see the whole workspace, and they choose how far every other role can see.

Admins can also scope a role to its reporting line, set a minimum group size, or switch individual metrics to pseudonyms.

Sample contributor metrics for the last 30 days, as the selected role sees them
Contributor Cycle time Review wait Merged
Asha K. you 1d 1h 2h 40m 14 Outside your scope. Not sent.
Mateo R. 22h 3h 05m 11 Outside your scope. Not sent.
Lin W. 1d 6h 1h 50m 9 Outside your scope. Not sent.
Priya S. 1d 9h 4h 10m 12 Outside your scope. Not sent.
Sam O. 2d 2h 3h 30m 7 Outside your scope. Not sent.
Noor A. 1d 4h 2h 55m 10 Outside your scope. Not sent.
Team aggregates, always visible
Platform 1d 2h 2h 32m 34
Payments 1d 16h 3h 32m 29

Sample data

What DevGlimpse reads

  • Pull request timestamps, sizes and states
  • Reviews and review comments, and when they happened
  • Commit counts and dates
  • Deployments, CI runs, tickets and incidents
  • Teams and org membership
  • AI tool usage, through an allowlist of fields. Anything else is dropped

What DevGlimpse never does

  • Store your source code, file contents or diffs
  • Record keystrokes, screen time, IDE activity or calendars
  • Rank individuals on a leaderboard
  • Keep your GitHub sign-in token. It is used once and dropped
  • Sell your data, or use it to train AI models

Read our measurement principles

Every number, defined in plain sight.

Black-box scores breed distrust. Every DevGlimpse metric has a published definition, one implementation, and tests with hand-computed answers.

  • Bots are excluded everywhere.
  • Sums travel with their counts, so any window averages correctly.
  • Teams and groups roll up from the same contributor rows.
  • Durations can run on calendar time or your working hours.
Read the full glossary
Cycle time

How long a pull request takes from opened to merged.

merged_at − created_at Counted on the merge day, for the author.
Review wait

How long the author waited for the first human review.

min(review.submitted_at) − created_at Bot reviews never start the clock.
PR size

How big the change was when it merged.

additions + deletions Smaller pull requests are reviewed faster and break less.
Throughput

The volume of shipped work, each counted on the day it happened.

commits · PRs opened · PRs merged · reviews Always the day of the event itself.

It comes to where your team already works.

A dashboard nobody opens changes nothing. DevGlimpse sends the signal to your chat, and answers questions from the assistant you already use.

An alert in Slack

DevGlimpse alerts · #eng-payments

Review wait is up for Payments. The median reached 6h 10m this week, against a target of 4h and a four-week average of 3h 30m.

Sample data

A question over MCP

Which teams' cycle time got worse this month, and is it review or CI?

Two teams. Payments went from 1d 4h to 1d 16h, almost all of it in the wait for first review. Mobile went from 22h to 1d 3h, and its end-to-end job is failing one run in six.

Sample exchange. The MCP server is read-only and answers inside the caller's own scope.

Alert rules
Targets, thresholds and anomalies on a team metric, sent by email, Slack or Microsoft Teams. Always the aggregate, never a person.
Weekly digests
A team digest, a personal one, and one for a channel. No dashboard visit required.
Your working week
Working hours, holidays and time zone, so a weekend never counts against review wait.
Dashboards your way
Saved views, per-team targets and read-only share links, plus a dashboard editor by plan.
Your org, not ours
An org hierarchy with director roll-ups, custom groups and contributor attributes.
Open by design
An MCP server for your AI assistant, and a REST API with an OpenAPI description by plan.

Trust you can verify, not a badge wall.

DevGlimpse is fully hosted, so there is nothing for you to run or patch. These guarantees are built in and tested today.

Walls between workspaces
Every query is scoped to your workspace by the GraphQL layer, and every table carries the workspace key, so a row can never point at another workspace’s data. Tested in CI with real tokens for different workspaces.
Least privilege, encrypted credentials
GitHub sync runs on short-lived installation tokens, and your sign-in token is dropped after sign-in. Access tokens and OAuth grants for other sources are stored encrypted under your workspace’s key.
Sessions that never leak
Fifteen-minute signed session tokens with rotating refresh. Sign-in hands over a one-time code, so a token never appears in a URL.
A key per workspace
Stored credentials and export files are encrypted under a key that belongs to one workspace. Deleting the workspace destroys the key, so copies left in a backup can no longer be read.
Every privileged action, on record
An append-only audit log covers role changes, approvals, invites, exports and more, with filters and a CSV export. Updates are refused at the database.
Your network, your rules
An IP allowlist restricts sign-in and every workspace request to your own ranges. Included by plan.
Your data, portable
Export the workspace as JSON whenever you like, or one person’s data on request. No ticket, no call.
Leave cleanly
Delete your workspace immediately, erase one person, or uninstall and get a visible 30-day read-only window before deletion.

58 capabilities are live. Here is what is not yet.

We'd rather be honest than impressive. Solid means it runs in production, a screen means it is being built, and an outline means it is designed and waiting. Design partners help decide the order.

LiveIn production today.

Building now In active development.

On the roadmap Designed, and next in line.

See every capability Read the changelog

Asked, answered.

Something else? [email protected]

What does DevGlimpse do?

DevGlimpse is AI impact and delivery analytics for engineering teams. It compares pull requests that carry an AI signal with the rest, per team: cycle time, review wait, PR size, reverts, and linked bugs and incidents. It computes DORA from your deploys and from PagerDuty, incident.io or Opsgenie. It reads metadata from the code hosts, trackers, incident tools and AI assistants you connect. Every formula is public.

How do you measure AI impact without ranking people?

From outcomes, not keystrokes. Commit and pull-request signatures classify AI-assisted work, and Claude Code, Codex CLI and Gemini CLI report usage over OpenTelemetry through a privacy allowlist. Results roll up per team and per tool. There are no individual rankings. Each developer can see their own AI usage privately. Individual numbers follow a scope your admin sets per role, and the server enforces it.

Does “no AI signal” mean the work was written by a human?

No. Inline completions leave no trace, so DevGlimpse never treats “no AI signal” as a human control group. Treat the AI numbers as a floor, and read the comparison as a comparison, not a cause. You can mark the day a team adopted a tool and read that team’s delivery before and after.

Does DevGlimpse read or store our code?

It never stores or shows source code. DevGlimpse reads pull-request, review, commit, deployment and membership metadata. On GitHub it needs Contents (read) only to list commits, and never fetches file contents or diffs. GitLab reports no line counts, so there a merge request’s diff is read once to count added and removed lines, and only those two numbers are kept.

Which tools are supported today?

Code: GitHub, GitLab and Bitbucket Cloud. Tickets: Linear, Jira Cloud and Asana. Incidents: PagerDuty, incident.io and Opsgenie. CI: GitHub Actions and GitLab pipelines. AI: Claude Code, Codex CLI and Gemini CLI over OpenTelemetry. Adapters for org-level Copilot usage, Cursor, and Anthropic and OpenAI usage are being built and are not in the product yet.

Who can install it, and how long does it take?

A GitHub organization admin signs in and installs the DevGlimpse GitHub App. A non-admin can copy an invite link to send to an admin. First numbers usually appear within minutes; the 90-day backfill runs inside the host’s rate limits and takes longer on very large organizations.

What does it cost?

DevGlimpse is free for up to 10 tracked contributors, with no time limit. A workspace with more than 10 active contributors can ask for a 14-day full-organization Team trial, then falls back to Free; nothing is deleted. Paid plans are priced per active contributor (a non-bot person with a commit, pull request or review in the last 30 days). Team has no minimum; Founding has a minimum of 20 paid contributors. Book a call or email us for terms.

Where is our data stored?

The service providers we use are listed on the sub-processors page, at devglimpse.com/subprocessors. We do not yet publish processing locations or a data-residency commitment; that page will carry them once they are confirmed.

What happens if we leave?

An admin can export everything the workspace holds at any time. Uninstalling the GitHub App from every connected organization puts the workspace in read-only mode with a visible deletion date 30 days out, and reinstalling before that date restores it. An admin can also delete everything immediately.

Do you have SSO or SOC 2?

Not yet. SSO with OIDC and SAML, SCIM provisioning and SOC 2 Type 2 are planned and listed on the public roadmap.

Build it with us.

Design partners: free until 31 January 2027, then a locked founding rate. Book 20 minutes.

Start free with GitHub Try the demo
Free to start
The Free plan tracks up to 10 contributors with no time limit, while we build together.
Founding pricing, locked
After the free period, the Founding rate, locked for 24 months.
A direct line to the builders
A shared channel with the engineers writing DevGlimpse, not a ticket queue.
Shape what ships next
Integrations, metrics and reports: your needs set the order.