Your agents are flying blind
An agent spins up in CI, fires off 500 model calls, and burns $3K overnight — and the only record is the invoice. Tokenoscopy is the flight recorder: every prompt, tool call, file edit, and dollar, replayable and capped.
Claude Code today · Codex, Cursor & Gemini CLI next · subscription and API-key fleets
A real session shape: an overnight backfill stopped by the budget guard at $24.81 — before the next $186.
Engineering
Replay, not archaeology
Rewind any session frame by frame — prompts, tool I/O, file diffs, subagents. Debugging drops from hours of log spelunking to minutes of scrubbing.
Finance
Spend with names on it
Cost per session, repo, branch, developer, and model — from local telemetry, not a month-late invoice. Subscription seats included.
Security
An audit trail that exists
Which repos did the agent touch? What ran? What got blocked, and why? Every action recorded with the decision that allowed it.
One command. No proxy.
No SDK rewrite.
Tokenoscopy rides Claude Code's own hooks and transcripts — your traffic never routes through us, and your agents never slow down.
01
Record
npx tokenoscopy init wires the hooks. Sessions stream in live; transcripts flush with secrets scrubbed before anything leaves the machine.
$ npx tokenoscopy init ✓ hooks installed ✓ redaction on ✓ recording
02
Replay
Every session becomes a shareable timeline: what the agent thought, ran, changed, and spent — down to the individual model call.
session 722ad2… 16 requests · $1.75 19 tool calls · +51 lines
03
Govern
Budgets per session, repo, or org. Alerts at 50/80/100%. And a guard that stops the next tool call when the cap is hit.
⛔ guard: deny Bash session cap $25 reached saved ~$186
Dashboards watch.
Tokenoscopy can stop.
Observability tells you what an agent spent after it spent it. Gateways cap tokens but can't see actions. The agent's own machine is the one place both signals exist.
The dashboard way
Trusted for charts, but the money is already gone.
Spend alert: $1,214
yesterday
Spend alert: $2,891
yesterday
Spend alert: $3,406
2 days ago
Invoice: $9,511
end of month
The Tokenoscopy way
Watch it live, replay it later, stop it in the act.
invoice-backfill · running in CI
$8.11 spent · 2,500 / 8,412 invoices
⛔ Guard: session cap $25 reached — Bash blocked
projected ~$186 not spent
Recommendation on the timeline
Batch verifications 50:1 → resume at ~$4 instead of $186.
Stopped. Before the next $186.
Per-session hard caps don't exist anywhere else — budgets are monthly, per-seat, or Enterprise-gated. Ours land mid-session, fail-open by design, and get recorded on the replay.
Caps so good, you'll let agents run overnight
Guard checks answer in milliseconds from precomputed state, fail open on any wobble, and never sit in your request path. The worst case is our outage — never your stalled session.
budget · every session
$24.81 spent · Bash blocked · saved ~$186
Cheaper than what it catches
You're paying for agent capacity either way — by the token or by the seat. Team pays for itself the first time it stops a runaway loop, or turns up a seat nobody's been using.
Free
$0 forever
- ✓Up to 3 developers
- ✓Full session replay
- ✓Spend attribution + weekly report
- ✓Budgets that alert — and show what they'd have stopped
- ✓7-day retention
Team
$99 /mo · first 10 seats
- ✓Budgets that actually block — per session, repo, or dev
- ✓Retry loops and idle seats surfaced weekly
- ✓Email & Slack alerts before a cap is hit
- ✓Per-repo, per-dev, per-model attribution
- ✓90-day retention · then $12/seat, falling to $6 at scale
Fleet
Custom
- ✓MDM rollout across every machine
- ✓SSO, audit export, org policies
- ✓Billed-vs-estimated reconciliation
- ✓Priority support
A seat is counted only when someone actually runs an agent that month — so the bill falls when usage does. Recording never stops, on any plan.
Stop reading invoices. Start reading flight logs.
We're onboarding design-partner teams running Claude Code in production now.