# Workboard by Cognitivi

> Workboard is one page for what frontier model coding tools cost a team: spend per call, session, project and machine, plan headroom, and a priced list of what to change. It is made by Cognitivi, a San Francisco company led by Roman Kagan, and costs $40 per seat per month for 10 seats or more. It is hosted by Cognitivi at app.cognitivi.ai or self-hosted in the customer's own AWS, GCP or Azure account. Prompts, replies, code and tool output never leave the developer's machine.

- What it shows: Overview (spend, savings available, active time, sessions, lines changed, plan headroom, spend by project, Needs you); Spend (by model, account, project, effort and speed, trend up to three years, budgets, cost reconciliation); Savings (a priced list of changes from eleven rules plus loops); Sessions (every session with its turns, agents, tools, cache misses, errors and compactions); Activity; Team (seats, invites, roles, admin visibility); Fleet (projects, jobs, pull requests, hosts, GPUs, sender health); Explore (any signal by any field, CSV download); Settings; org reports from the Anthropic Admin API.
- Why it saves: on a measured account over 30 days, 96% of tokens were context re-read on every call (about 177,000 tokens per call), and Workboard's rules priced about 18% of spend ($1,420 of $7,760) as avoidable: contexts over 200,000 tokens, agents idle past their five-minute cache, high effort on tool-only turns, unused MCP servers, idle compactions, cold model switches, permission waits, tool failures, cache misses and retries. Applied to a $50 million annual bill, 18% is about $9 million a year, against under $1 million of seats for 2,000 developers.
- How it works: one Rust sender (wb) per machine for macOS, Linux and Windows, installed with one line and joined with one browser approval; four channels (OpenTelemetry export, hooks, a 60-second pass, the status line); every event cut to a fixed list of fields on the machine and checked again on the server; idempotent writes; a seven-day outbox; totals kept three years. Workboard makes no model calls of its own.
- Security: prompt text, replies, file contents, code, diffs, tool output, paths and arguments never leave the machine; a canary string in the tests must never appear in any output; https only; one token per machine stored as a SHA-256 hash and revocable alone; members see their own data, sharing is opt-in, admin visibility is an owner-only disclosed switch; nightly encrypted backups with a monthly restore check; no certification yet, SOC 2 Type I planned; data never used to train models.
- Models: Claude Code today. Next: OpenAI Codex CLI, then Google Gemini CLI, OpenAI Platform, GitHub Copilot, Cursor, xAI Grok.
- Products: Workboard (live); Vault (in design: placeholder tokens instead of PHI, PII and PCI values, de-anonymised on the customer's side); TokenGuard (coming: budgets, limits and a managed policy pack for frontier model spend); Advisor (coming: only the priced recommendations, by email or chat); Agenda (coming: the outstanding and recommended tasks agents leave behind, as one roadmap). Names are working names.
- Pricing: the Team plan, hosted by Cognitivi, is $40 per seat per month with every feature included, from 10 seats; machines take no seat; checkout by Polar. The Enterprise plan runs in the customer's own cloud account and is priced with the customer.
- Forward deployed engineering: an engineer who runs Workboard on their own work sets up the senders and the self-hosted server, reads the numbers with the team and fixes what costs the most. Today that engineer is Roman Kagan.

## Pages

- [Home](https://cognitivi.ai/): what Workboard shows, the $50 million illustration, security summary, how it works, products, coverage, deployment and pricing summary.
- [Products](https://cognitivi.ai/products.html): Workboard, Vault, TokenGuard, Advisor and Agenda.
- [Why it saves](https://cognitivi.ai/savings.html): where the tokens go, the eleven rules, the four levers, a savings calculator, and the rules on the roadmap.
- [Pricing](https://cognitivi.ai/pricing.html): $40 per seat per month from 10 seats, the Enterprise plan, and common questions.
- [Security and privacy](https://cognitivi.ai/security.html): what leaves a developer's machine, the controls in place, the questions security teams ask, and the controls being rolled out.
- [How it works](https://cognitivi.ai/platform.html): the sender, the four channels, the path of one event, OpenTelemetry, data tiers, the access model and the path to scale.
- [Enterprise deployment](https://cognitivi.ai/enterprise.html): running Workboard in your own GCP, AWS or Azure account, what runs where, the buyer's six questions, and forward deployed engineering.
- [Request a demo](https://cognitivi.ai/request-demo.html): the form to request a walkthrough on your own numbers.
- [Contact](https://cognitivi.ai/contact-us.php): contact form, email rkagan@cognitivi.ai, phone +1 (415) 426-8280, and the office at Cognitivi, 4444 Geary Blvd #202, San Francisco, CA 94118.

## Optional

- [Full text](https://cognitivi.ai/llms-full.txt): every public page as plain Markdown.
- [Privacy Policy](https://cognitivi.ai/privacy-policy.html): how Cognitivi handles personal information on cognitivi.ai.
- [Terms of Service](https://cognitivi.ai/terms-of-service.html): terms for the cognitivi.ai website and Workboard.