Specivo shows how much each person on your team spends on AI, and on which project, feature and issue. Start at the company total and open it down to a single agent session. - Collected automatically from Claude Code, Codex, Gemini CLI and OpenCode - Every session tied to the issue it worked on - Runs on your own server; only usage numbers leave the laptop ## The problem Teams use Claude Code, Codex, Cursor and other AI tools every day, on personal plans, company seats and API keys. The bills come from different places, and none of them says who spent the money or on which feature. - **Who on the team spends the most on AI?** Spend per person, with the tools each one uses and how that usage is paid for. - **What did this feature cost in AI?** Every usage entry is attached to the issue the agent was working on, so each feature, bug or task carries its own AI cost. - **Which tool or model is driving the bill?** Group the same spend by tool, model, provider or cost type, and open any row to see the entries behind it. ## From company to session Open a project, then a person, then a feature, and finally the sessions behind it: date, tool, model, tokens and spend. Switch the grouping at any level. The page shows this as an interactive demo report for a fictional company, ACME Waste & Recycling, over the last three months. - Reports: an admin report for the whole company, a spend report per project, an AI usage tab per project, AI usage on each issue, and "My AI usage" for every person. - Group by person, project, issue, tool, model, provider, cost type or git branch, with a second "then by" grouping. - Filters for person, tool, model, cost type and dates, and a link from every row to the underlying entries. - Each entry shows date, person, tool, model, tokens, cost type, spend, project, issue and branch. ## Cost per feature Every usage entry is attached to the issue (feature, bug or task) the agent was working on. Specivo works it out from: 1. The agent's own Specivo tool calls 2. An environment variable, `SPECIVO_ISSUE` 3. Commits 4. Branch names 5. The repository Long sessions that switch between tasks are split per issue. Research done before the issue existed still counts towards that issue. Entries someone has moved by hand are never overwritten. Usage with no issue lands on the project, or as unassigned, where people can move it. ## What did this feature cost? Open the issue. Every issue has an AI usage tab. It adds up the agent sessions behind the feature, its sub-issues included: spend, API-equivalent value, tokens, the people who worked on it, the tools and models they used, and how the cost built up week by week. Spend is shown by person, by tool and by model; every session behind it is listed page by page; and you see only what your project role allows. ## Every person, every tool See what each person spends on AI, which tools they use, and how that usage is paid for: a personal plan, a company seat or pay-per-use API keys. - Everyone sees their own usage on "My AI usage". - People see the AI usage their project role allows. - Individual subscription costs are shown only to those allowed to see them. ## How it works 1. **Install the reporter.** `ai-usage` is an open-source command-line tool (MIT, Python) for Windows, macOS and Linux. It reads the local logs of Claude Code, Codex, Gemini CLI and OpenCode. 2. **Work as usual.** Its hooks report usage after each agent turn. Nobody has to log anything. 3. **Open the reports.** Company, project, issue and personal views, down to every entry. ```bash ai-usage init ai-usage hooks install --tool all ``` ## Charts - Spend over time, stacked by person, tool, model or project, with automatic day, week or month grouping - Tokens over time: input and output versus cache - Breakdowns by person, project, model and tool - The data is also available as CSV, through the REST API and through MCP ## All kinds of cost Spend is pay-per-use API charges plus each subscription's fee, spread over the usage that subscription covered. Cost types: subscription, API (pay-per-use), local, free. - **Personal plans:** presets for Claude Pro and Max, ChatGPT Plus and Pro, Cursor Pro, GitHub Copilot, Google AI Pro and Ultra, and SuperGrok. - **Team seats and custom plans:** business plans such as team seats, and any custom plan. Any currency, with exchange rates. - **Model prices:** a bundled list of prices for more than 300 models, with history. Admins can override any price. Usage without a price is flagged, never silently counted as zero. Two extra figures sit next to spend: API-equivalent value, the same tokens at pay-per-use prices (an estimate, not a bill), and leverage, that value divided by the plan fee. ## Privacy and self-hosting The reporter sends only usage numbers (tokens per model, per session and per day) to your own Specivo server. It never sends prompts, code or file contents. It is open source (MIT), and its API key is stored in the OS keychain. Specivo runs on your server, air-gapped installs included. ### Every piece is open source Anyone can read what is collected, what is sent and where it goes: - [Specivo server](https://github.com/specivo/specivo) (AGPL-3.0): stores the usage and builds the reports, on your own server. - [ai-usage reporter](https://github.com/specivo/ai-usage-reporter) (MIT): read exactly what it collects and sends. The ingest contract, the JSON format of what is sent, ships with it. - [Agent skill](https://github.com/specivo/specivo-agent-skills) (MIT): lets Claude Code, Codex and Cursor report their own usage. ## Agent-native Specivo's MCP server and an agent skill for Claude Code, Codex and Cursor let agents report their own usage. One line in `CLAUDE.md` or `AGENTS.md` turns it on. You can ask your agent to register your subscriptions from your invoices. ## Get started - [See the live demo](https://specivo.io/live-demo/) - [Self-host in 15 minutes](https://specivo.io/self-host-specivo-docker/)