Tokenbase turns the session logs your agents already write to disk into spend and usage by team.
curl -fsSL https://tokenbasehq.com/install.sh | shTokenbar for macOS
Hover the top edge for a compact status island. Click once for active sessions, provider limits, and the local tools around the work.
macOS 14+ · Apple silicon and Intel · signed and notarized
Auto-hide leaves a small top-edge handle. Hover reveals status after 250 ms; clicking keeps the full island open.
Tokenbar
2 working · 1 recent
Resting
A 48 px handle
Hover
Status in 250 ms
Click
Command surface
Working and recent agents share one view. Attention requests, todos, tools, and bounded transcript detail stay attached to the session that produced them.
Tokenbar
2 working · 1 recent
WORKING
tokenbase
CodexAttentionThis Mac · Desktop · Working
Waiting for your answer
LAST REPLY
The island interaction is ready for the final visual pass.
TODOS
Wire hover presentation
Verify reduced motion
menubar
Claude · CLI · Working
See provider windows, reset times, pace, projected exhaustion, and estimated local cost without leaving the island.
Provider limits
Real SwiftUI render · example data
Open, copy, or safely stop servers owned by your user. Jump to project instructions, manifests, agent config, skills, plugins, and logs.
Servers
Listening on this MacVite :4173
tokenbase · node · Up 18m
Next.js :3000
sample-app · node · Up 1h 12m
Python :8000
ingest-api · server · Up 42m
Routes
Local onlytokenbase · Project
Codex · Global
Save explicit shell actions, launch terminal workspaces, sort sessions on the Agent Board, and pin the detail you need as a widget.
Fast Actions
Full local check
bun run ci
Workspace
3 panesAgent Board
Example stateTranscript · tokenbase
Keep transcript, tools, todos, or attention in view while the agent works.
Local boundary
Tokenbar reads supported local data during ordinary monitoring. It does not turn transcript content into a second telemetry stream.
Built for the Mac you already develop on. No Tokenbase account is required.
Your team runs AI coding agents all day, and almost none of how they work is visible.
The usage data already sits on every developer's machine, scattered across agents. Tokenbase pulls it together and turns it into comparable cost, activity, and workflow signals that show where to investigate.
One view across every coding agent your teams run, rolled up by team, never down to a person.
Claude Code, Codex, Copilot, and Cursor, normalized to one format so the spend actually compares.
| Team | Sessions | Spend |
|---|---|---|
| Platform | 412 | $3.1k |
| Web | 318 | $2.4k |
| Mobile | 247 | $1.9k |
| Data | 203 | $1.6k |
Every dollar rolls up to the team that ran it. Cohorts too small to stay anonymous stay hidden.
Cost per session for each agent, with unavailable costs shown as n/a instead of forced into the comparison.
3 sessions cut off by the plan this week
Measured from the agents' own rate-limit telemetry: how hard each plan window is being used, and the sessions your subscription cut off mid-work.
An overnight run lands on the hours it worked, not the day it started.
Every number is read from the agents' own logs and attributed to the time the tokens actually flowed. Verified adapter by adapter against the tools' own source.
Never leaves the machine.
Built from usage metadata alone. Source, prompts, and file contents never leave the machine.
Tokenbase surfaces measurable usage patterns, starting with prompt cache reuse, and gives each team a concrete place to look next.
So Tokenbase marks every number as one of two things, and never dresses up a guess as a fact.
Facts. Tokens, models, sessions, and tool calls, straight from the logs. Cost is computed from them.
Estimates. Retries, abandoned sessions, likely task type, inferred from the data, and always labeled as such.
A small collector reads the logs on each machine. Run the analysis in the cloud, your own VPC, or fully on-device, your call.
Reads the logs your agents already write. Read-only.
Every agent mapped to one format, so the numbers compare.
2 small teams hidden · cohort under 5
Rolls up to team totals. Small teams stay hidden.
Every insight is built from usage metadata, never your code or prompts. By default that is all that leaves the machine. Pick the setup that fits your security review.
| Setup | What leaves the machine | Analysis runs | Best for |
|---|---|---|---|
| SaaS · metadata-only | Structured metadata only | Cloud (no transcript) | SMB / midmarket |
| SaaS · redacted text | Redacted excerpts, opt-in | Cloud or customer LLM key | Less sensitive orgs |
| VPC / on-prem | Configurable | Customer VPC / local model | Enterprise / security |
| Local-only pilot | Nothing | Local model or no LLM | Security evaluation |
What never leaves the machine
Pricing
Free for individual developers and any team up to three. Past that, one flat price for the AI spend you put under management. Adding a teammate, tool, or session never costs more.
On-prem collector fleet, custom terms.
Prices in USD, billed monthly; annual billing saves about two months. Spend bands are soft: cross one and we prompt an upgrade, never cut off your data.
Every AI coding agent your team runs, in one view. Start with a metadata-only pilot.
%20--%3e%3cdefs%3e%3cstyle%3e%20.st0%20{%20fill:%20%23edecec;%20}%20%3c/style%3e%3c/defs%3e%3cpath%20class='st0'%20d='M457.43,125.94L244.42,2.96c-6.84-3.95-15.28-3.95-22.12,0L9.3,125.94c-5.75,3.32-9.3,9.46-9.3,16.11v247.99c0,6.65,3.55,12.79,9.3,16.11l213.01,122.98c6.84,3.95,15.28,3.95,22.12,0l213.01-122.98c5.75-3.32,9.3-9.46,9.3-16.11v-247.99c0-6.65-3.55-12.79-9.3-16.11h-.01ZM444.05,151.99l-205.63,356.16c-1.39,2.4-5.06,1.42-5.06-1.36v-233.21c0-4.66-2.49-8.97-6.53-11.31L24.87,145.67c-2.4-1.39-1.42-5.06,1.36-5.06h411.26c5.84,0,9.49,6.33,6.57,11.39h-.01Z'/%3e%3c/svg%3e)
