Speaks OpenTelemetry
Workboard takes in Claude Code's OpenTelemetry export over OTLP: logs and metrics, sent as OTLP/JSON over https, next to the batches from its own collector.
Workboard, by Cognitivi
Illustration with example data.
See what Claude Code costs your team, and where the time goes.
Workboard by Cognitivi shows what Claude Code costs your team: usage and cost for every account, machine and project on one page, with each session down to its turns and agents. Prompts, replies and code stay on your developers' machines.
$40 per seat per month, from 10 seats. Hosted by Cognitivi or run in your own cloud.
Product
One page for every account, machine and project
A small collector on each machine reads Claude Code's own records and sends counts, costs and names to one server. Workboard makes no model calls of its own.
Org reports: for API and Team or Enterprise organizations, Workboard pulls the Anthropic Admin API every hour: daily usage and cost by model and the Claude Code report per user.
01
A collector on each developer's machine, one server for the whole team.
alex - dev-mac-1
priya - dev-mac-3
ci-runner-2
The collector reads Claude Code's own records. Prompts and code stay here.
counts, costs, nameshttps only
Workboard serverHosted by Cognitivi, or in your AWS, GCP or Azure account
One pageEvery account, machine and project
How it works
Built on OpenTelemetry
Workboard's collector and Claude Code's own OpenTelemetry export both feed one Workboard server. The data model behind the page is moving to OpenTelemetry's ids and names.
One data model, aligned with OpenTelemetry
- An ontology: one shared definition each of org, user, machine, account, project, session, turn, API call, tool call, agent, job and loop, used the same way by the page, the API and the export.
- Trace and span ids per session, so a session reads as one trace of its turns, API calls, tool calls and agents, with OpenTelemetry's GenAI semantic names.
- Export of spans and metrics over OTLP to your own observability platform, such as Datadog, Grafana, Honeycomb or New Relic.
- A Prometheus
/metricsendpoint with totals per org, user, project and model.
Security and privacy
Your code stays on your machines
On our hosted service the collector sends counts, sizes, times, costs and names to Cognitivi's server. Self-hosted in your own AWS, GCP or Azure account, the server is yours and nothing leaves your network. Either way, prompts, replies and code stay on the developer's machine.
Stays on the developer's machine
- Prompt text and Claude's replies
- File contents, code and diffs
- Tool output
- Request captures, when a developer records one
Sent to the Workboard server
- Counts and sizes: tokens, prompt length, lines changed
- Times and costs of each call, tool and session
- Names: model, project, tool, host, and short tool labels
- Titles of sessions and pull requests
- Over https only: a sender refuses a plain http address
Models
Claude Code today, other frontier model tools next
Workboard reads Claude Code now. Support for the coding tools of other model providers is coming, so one page can show all of them side by side.
Available
Claude Code
Sessions, agents, tools, plan windows, OpenTelemetry and the Anthropic Admin API.
- OpenAI Codex CLI
- Google Gemini CLI
- GitHub Copilot
- Cursor
- xAI Grok
- Others
Deployment
Hosted by us, or run in your own cloud
On Cognitivi's servers
We run the server. Your developers install the collector with a sender token and their data shows up within a minute. Counts, costs, names and titles reach our server; prompts, replies and code stay on the machines. Each organization sees only its own rows.
In your GCP, AWS or Azure account
The server runs in your cloud account with the cloud's managed PostgreSQL and your own identity provider. The same collector sends to your server instead of ours, so nothing leaves your network.
Forward deployed engineering
An engineer who uses Workboard, working with your team
An engineer who runs Workboard on their own Claude Code work sits with your team, sets up the collectors and the self-hosted server, reads the numbers with you and fixes what costs the most. Today that engineer is Roman Kagan, who leads Cognitivi.
Pricing
$40 per seat per month
From 10 seats, any number above. Every feature is included. Self-hosted installs are priced with you.
- 10 seats
- $400
- a month
- 15 seats
- $600
- a month
- Self-hosted
- Quoted
- per install
Company
Built by Cognitivi
Cognitivi is a San Francisco company led by Roman Kagan, with a background in search relevance, information retrieval and evaluation. Workboard grew out of measuring our own agent work across many machines and projects: what it cost, where it looped and what was waiting on a person.
See your own numbers in a walkthrough
Tell us how many developers use Claude Code and where you would run Workboard. A person at Cognitivi reads every message and replies.