AI agent time and cost trackingBill agent work with time and LLM costs accounted for
Capture agent sessions against the right project, record measured model usage as an expense, and review both alongside your team’s work before invoicing.
Map a repository or workflow once. Completed sessions arrive on the right client, project and task with the duration and delivery context intact.
Track configured Codex and Claude Code sessions with the Agent Skill
Use the CLI, SDKs or API for custom agent workflows
Map each repository to the right client and project
Agent sessions
Today · 3 completed
Source: Agent
Claude Code
Draft
Northstar website · Implementation
feature/invoice-review
1:42:18
Codex
Draft
Atlas rollout · QA
fix/onboarding-tests
0:48:32
Research agent
Draft
Northstar website · Research
session 4f8a
0:36:04
Each repository uses its own client, project and task mapping
Know what agent work costs
Keep agent time and measured model usage beside the human work on the project. Filter by source and see what each delivery actually cost.
Record measured model usage as an LLM expense
Compare agent and human work with source filters
Keep session and model context available for review
Agent delivery
Northstar website · October
Export
Agent time
12:36
LLM usage
$86.40
Billable value
$1,890
WorkSourceAmount
Implementation
Claude Code · 8:12
Agent$1,230
QA and research
Codex + research agent · 4:24
Agent$660
LLM Usage
Measured provider usage
Expense$86.40
Bring agent work into the invoice
Add approved agent time and eligible model expenses to the same invoice as your team’s work. Keep the client-facing detail clear while internal metadata stays internal.
Set billable and cost rates for agent work
Group invoice lines by project, person or task
Keep internal session metadata off the invoice
Draft invoice
Northstar website · October
Draft
Strategy and review
Maya Chen · 6:00
$1,200
Implementation
Claude Code · 8:12 · Agent
$1,230
QA and research
Codex + research agent · 4:24
$660
LLM Usage
Measured provider usage
$86.40
Grouped by task · Internal metadata hidden
Invoice total
$3,176.40
Talk to Keito from your AI assistant
Review project data, control timers, and prepare billing through a connected assistant
The tools to capture agent delivery, understand its cost and carry reviewed work into client billing
Capture agent work
Agent Skill for Codex and Claude Code
Repository-specific project mapping
Completed session tracking with the CLI
Node and Python SDK integrations
LangChain and CrewAI patterns
REST API for custom agents
Keep the full context
Agent source labels on time and expenses
Client, project and task attribution
Session, model, branch and revision metadata
Measured LLM usage as project expenses
Source filters across reports
Human and agent work in one project record
Review and bill
Draft entries ready for human review
Billable and cost rates for agent work
Agent time in approval workflows
Invoice grouping by project, person or task
Claude hosted MCP connector available
ChatGPT direct connector in developer beta
Questions about AI agent tracking
What is AI agent time and cost tracking?
AI agent tracking records the sessions, time and measured usage costs created when an agent works. Keito connects those records to the same clients, projects and tasks as your team’s work, so you can review delivery, understand costs and prepare accurate billing from one project record.
How does Keito track agent work automatically?
For Codex and Claude Code, the Keito Agent Skill installs local hooks and uses a repository-specific configuration to record completed sessions. Custom agents can use the CLI, REST API, Node SDK or Python SDK to start and stop sessions and log measured LLM usage. Keito only records repositories and workflows you configure.
Does connecting Claude or ChatGPT automatically track every conversation?
No. The hosted MCP connector and automatic agent tracking are separate. Claude or ChatGPT calls a Keito tool only when you ask it to in a conversation. It does not monitor chats, track time in the background or measure the model usage of the conversation for you.
Which AI agents can I track with Keito?
The Keito Agent Skill supports configured Codex and Claude Code sessions. The CLI, REST API and SDKs can record work from custom agents and frameworks such as LangChain and CrewAI. Every integration can attach the agent, session, model and project context needed for reporting.
Can I use Keito from Claude?
Yes. Keito’s hosted MCP connector is available for Claude as a custom remote connector. After you authorize your Keito account, Claude can use the Keito tools allowed by your role—for example, finding projects, controlling timers, reviewing time and expenses, or preparing billing. Write actions remain visible in the conversation.
Can I connect Keito directly to ChatGPT?
Keito’s direct ChatGPT MCP connection is currently in developer beta and is not yet listed in ChatGPT’s public directory. Eligible users can connect through ChatGPT developer mode. The connector uses Keito OAuth and follows the permissions of the signed-in user and selected workspace.
Can agent work be included on a client invoice?
Yes. Agent time entries and measured LLM expenses can be reviewed with human work and imported into an invoice. Group invoice lines by project, person or task, set the appropriate rates and use clear notes so the client sees useful delivery detail rather than internal session metadata.
Does Keito store agent prompts or code?
Agent time entries do not require Keito to store prompts or code. The Agent Skill sends the configured session record and metadata to Keito, and you can enable note redaction so transcript-derived notes are replaced with generic notes. Do not put sensitive information in entry notes or metadata.
Do AI agents require additional Keito seats?
No. Every Keito plan includes unlimited AI agents. Solo includes one licensed human user; Pro and Business include unlimited licensed users as well as unlimited agents.
Give agent work a place in every project
Start with one configured workflow. Keep the sessions, costs and client context together from the first run to the final invoice.
