Who did what, why, and how much it cost in AI.
You ask in Claude, via the MCP connector, and the answer comes with the events that prove it.
And nobody on the team fills out a form, moves a card, or writes a report for that to happen. The work is the data.
-
Claude Code -
Cursor -
Codex -
OpenCode -
Kiro -
GitHub -
GitLab -
Linear -
Jira -
Monday.com
Who this product is for
You stop asking “how are we doing?”
Team status stops depending on standups, spreadsheets, or someone remembering to move a card. You ask in natural language and get the answer with the real history behind it: who did it, when, and what happened along the way.
- The full timeline of a delivery, from plan to merge, including what got stuck.
- How long each pull request spent in review and where the queue is stuck right now.
- What shipped in the week by repository, branch, or person, with the events that prove it.
- Agent usage patterns: which dev uses AI more effectively, and why.
How it works
Everything your team already produces enters the graph with a timestamp and comes out through MCP, however you ask.
- Claude Code
- Cursor
- Codex
- OpenCode
- Kiro
- GitHub Webhook
- GitLab Webhook
- Jira
- Linear
- GitHub Projects
- Monday
Ask in natural language from any MCP-compatible client. No SQL, no BI, no waiting on anyone's report.
- AI cost per feature, PR, and person
- Review cycle: where the queue gets stuck
- Complete timeline of a delivery
- Decisions and trade-offs preserved
Questions a team already asks today
Real answers from a production account, with names of people, repositories, and partners removed. Each one comes with the events that back it.
What did the team ship this week?
Heavy week. 218 pull requests merged and 985 commits, against 36 open and 19 closed without merging. Code balance was 153k lines added against 64k removed, which points to new construction rather than refactoring.
Activity isn't evenly spread: the three busiest repositories accounted for 60k of the 153k lines, almost 40% of the total. The rest is spread across another 38 repositories with low volume.
- Pull requests
- 218 merged · 36 open · 19 closed
- Commits
- 985
- Code
- +153,535 / −64,482 lines
- People
- 74 active devs across 41 repositories
- AI cost
- $17,180.70 for the week
Aug 14–21 · 41 repositories · cost measured per AI session, not allocated
Which pull request is stuck waiting for review?
There are 49 open right now, and the queue has two different problems. The oldest has been stuck since July 23, 29 days, with four files changed and no review comment to date. It isn't a big PR, it's a forgotten one.
The costliest case is different: a PR opened August 7 with 46 files, 7,820 lines, and 32 commits. It received eight review comments and still hasn't moved in two weeks. And there's a third, flagged high-priority, with 36 files changed and zero comments since August 11.
- Current queue
- 49 open pull requests
- Oldest
- 29 days stuck · zero comments
- Biggest stuck
- 7,820 lines · 46 files · 14 days
- No review at all
- 3 of the high-priority PRs
queried Aug 21 · 30-day window · sorted by inactivity
Why is shipping baked into the product price?
Someone already answered that. Back in June, a dev spent a whole AI session chasing the origin of that rule, digging through old code and discussions. The session closed, the conclusion never turned into a document, and it's still queryable today, with the full reasoning, not just the verdict.
What the investigation found: the payment platform never knew how to separate shipping, so the store always sent the full amount, product plus shipping, to payment and invoicing. For the order total to add up, a rule was added that spreads shipping across each item's unit price, which inflates the shelf price. Whoever implemented it noted at the time that it was a stopgap. No partner ever adopted separate shipping, two tried and gave up, so it was never revisited.
- Source of the answer
- AI session from June 17
- Root cause
- the payment platform doesn't separate shipping
- Stopgap
- shipping spread across each item's unit price
- Why it stayed
- no partner adopted separate shipping
AI session from June 17, 2026 · the reasoning was preserved, not just the conclusion
How much did this pull request cost in AI?
$74.05, across 1h30 of AI session spread over eight episodes, all on Opus 5 and by a single dev. The number comes from the tokens actually consumed in that session, not from allocating the monthly bill across seats.
This level of granularity is what enables the next question, usually the one that matters: was the whole feature worth what it cost? That same week, the 74 devs consumed $17.1k, an average of $232 per dev. Knowing the average doesn't help anyone decide anything. Knowing a specific PR cost $74, it does.
- PR cost
- $74.05
- AI session
- 1h30 · 8 episodes · 1 dev
- Model
- Opus 5
- Week's context
- $17.1k / 74 devs
real session tokens tied to the pull request they produced
Pricing
The plan is sized by the volume of events your team generates per month. Use the slider to estimate yours.
No seats and no billed queries: volume ingested decides the bill, not how many people open the client.
What counts as an event
An event is an action coming from your connected sources. Prose (a PR description, a review comment, an issue, an AI coding session) goes through a language model to be understood and counts as 1 event. Metadata (a push, a CI run, a deploy) is recorded without an LLM and counts as 1/5 of an event.
Example: a review comment explaining why the approach changed counts as 1 event. Five pushes on the same day add up to 1 event.
- GitHub
- Pushes, pull requests, reviews, issues, comments, and deployments.
- Linear
- Issues, comments, projects, and project updates.
- Jira
- Issues, comments, status transitions, and sprints.
- Monday.com
- Board items, updates, and column changes.
- Claude Code, Cursor, Codex, OpenCode, and Kiro
- AI coding sessions. Event breakdown varies by tool.
What if the team goes over the plan?
Nothing stops being ingested. The overage goes into the next invoice as overage , in packs of 1,000 events for US$ 49 per pack.
- Real-time consumption in the portal. You track the ongoing cycle and see the overage building up, not discover it on the invoice.
- Overage can be switched off per cycle. A toggle in the portal caps the account at the contracted volume for months you don't want overage.
- Event blacklist. Event types your team doesn't care about are blocked before they come in and never count toward volume. Just add them to the list with one click.
Cogniscape and DX
DX asks developers how it went. Cogniscape reads what happened.
DX measures developer experience and productivity, combining surveys with system metrics. Cogniscape captures the work itself, passively. They are different approaches and partly complementary: what changes is what each one sees.
Core question
- DX
- How are developer productivity and experience doing? (DX Core 4, DXI, benchmarks)
- Cogniscape
- Who did what, when, how, why, and what did it cost in AI?
Data source
- DX
- Surveys (DevSat, targeted studies, experience sampling) and SDLC metrics
- Cogniscape
- Passive capture of the work: hooks in AI agents, repository and project-management webhooks
Team effort
- DX
- Developers answer periodic surveys
- Cogniscape
- Zero: install the CLI once and forget it. No surveys, no new dashboard
AI cost
- DX
- Measures AI usage, impact and ROI (AI-generated code per commit, PR, team, agent, repo)
- Cogniscape
- Real token cost attributed to PR, feature and person, not a monthly allocation
Context and decisions
- DX
- Focus on metrics and perception
- Cogniscape
- Preserves the context of the work: timeline from planning to merge, blockers, decisions and trade-offs behind the diff
How you query it
- DX
- Platform dashboards and reports
- Cogniscape
- Natural language over MCP, right inside Claude, Cursor or any MCP client, no SQL or BI
Sources
- DX
- Data Connectors (list not disclosed on the public page)
- Cogniscape
- Claude Code, Cursor, Codex, OpenCode, Kiro; GitHub, GitLab; Jira, Linear, GitHub Projects, Monday.com
Audience
- DX
- Enterprise engineering organizations (customers named on the site: Adyen, Dropbox, Vanguard, Booking.com)
- Cogniscape
- Engineering managers, CTOs, CFOs and founders; teams of dozens of devs
Pricing
- DX
- Per developer, on request (not published)
- Cogniscape
- By event volume, no per-seat or per-query pricing; 15-day pilot, no card. See the pricing section
Data and privacy
- DX
- —
- Cogniscape
- Hosted in the European Union, under LGPD and GDPR; the developer's prompt text is never captured
When to use DX
To measure perception and productivity maturity, with market benchmarks.
When to use Cogniscape
To see the real work, the AI cost per delivery, and query it all by conversation.
Many companies use both views: what the team feels and what the team did.
Does Cogniscape replace DX?
Not necessarily. DX focuses on developer experience and productivity, with surveys and benchmarks. Cogniscape captures the work itself and answers who did what, when, how, why and what it cost in AI. Many companies use both views, and the choice depends on the question you need answered.
What is the difference between measuring perception and capturing the work?
Measuring perception means asking developers how it went, through surveys, and combining that with system metrics. Capturing the work means passively recording what happened, with no survey: AI sessions, pull requests, issues and decisions, with dates and token cost. The first shows how the team feels; the second, what was done and what it cost.
Comparison based on public information from getdx.com as of October 2026. Trademarks belong to their respective owners.
Who already uses it
Cogniscape has become an essential observability layer for our engineering team.
At Juntos Somos Mais, we needed to centralize and speed up the analysis of the metrics that actually matter to how we work, and that's exactly what Cogniscape delivers.
What sets the partnership apart is response speed: new connectors ship fast, the team is genuinely close, and the product evolves at the pace our operation needs.
Murilo Cezar Amêndola de Oliveira Engineering Leader, Juntos Somos Mais With Cogniscape, I stopped being the annoying manager who keeps asking the engineering team “how are we doing?”
Having a tool that works transparently and adds no friction to the development process helps us guarantee full adoption by the team.
Through MCP, we have full power to extract and cross-reference data, generating information that's genuinely useful for confident strategic decisions.
Julio Bitencourt Head of Software Engineering, Qeevo Group Bring a question nobody can answer today
In 15 minutes we show the answer coming straight out of a real team's graph. Then we connect yours: GitHub, project management, and AI sessions. 15-day pilot, no card required.