Flight recorder + spend governor for AI agents
Flightbox runs your agent pipelines through a managed runway. Deterministic Python owns the loop — so every tool call, prompt, and dollar is on the record, and a run that blows its budget gets stopped, not billed.
A retry loop here, an oversized context there. Nothing fails, nothing alerts — and the invoice climbs while nobody is watching.
Observability SDKs watch from the sidelines. They can tell you what happened. They cannot make it stop happening.
Most teams meet theirs at full speed, at 2 a.m., in a loop no one is reading.
A hard budget gate suspends the run before the next dollar — the crossing call is billed, recorded, and nothing runs past it. A stop, not an alert.
Pricing
Recording, replay, budget caps and the hard mid-run halt are all in the open-source core — yours today, at no cost, on every install. Paid plans are founding-member places in Flightbox Cloud: the layer that proves the policy held, across your whole org, after the fact. Priced by the agent spend you put under governance, never per seat.
Read this before you pay. Flightbox Cloud has not shipped yet. A founding membership funds its development and locks your rate for the life of your subscription — it does not unlock anything in the core, because nothing in the core is locked. The halt that stops a runaway agent is free and always will be. Cancel any time.
$0
OPEN SOURCE · SELF-HOSTED · NO LIMIT
$29
PER MONTH · $290/YR · RATE LOCKED
Governs up to $1,500/mo of agent spend
$149
PER MONTH · $1,490/YR · RATE LOCKED
Governs up to $10,000/mo · unlimited seats
$499
PER MONTH · $4,990/YR · RATE LOCKED
Governs up to $40,000/mo · unlimited seats
Self-hosted control plane, unlimited governed spend, SCIM, custom retention, SOC 2 posture and a support SLA — scoped with you, billed annually. Also in development; talk to me early and you shape it.
NOT YET AVAILABLE — SHIPS WITH FLIGHTBOX CLOUD
No charge, no account — this opens your mail app so you send it yourself, which means I never collect anything you didn't choose to send. One reply when Cloud opens, and nothing else.
14 DAYS BEFORE YOUR FIRST CHARGE · CANCEL ANY TIME · RATE LOCKED FOR THE LIFE OF YOUR SUBSCRIPTION
How it works
Flightbox is a managed runway, not a sidecar. Deterministic Python owns the loop; your agents are bounded nodes inside it — which is why it can stop them, not just watch them.
STEP 01
Install the OSS core into your repo. Define your pipeline — phases, agents, budgets, approval points — in plain config. Run it with just.
STEP 02
Every tool call, prompt, and decision streams live into a SQLite trace. Budget and approval gates sit in the loop itself — when a cap is crossed or an approval times out, the run halts. Hard stop, not a warning.
STEP 03
Open the session-replay UI: a waterfall per run, per-phase tool-call detail, per-agent dollar attribution. Grant pending approvals from any terminal. Get cost estimates before the next run buys a single token.
Features
Every tool call, prompt, and decision streams live into a SQLite trace. Session-replay UI with a waterfall per run and per-phase tool-call detail — exactly what your agents did, on the record.
Per-call token and dollar breakdown, rolled up per agent, per phase, per run. Know that the reviewer costs 3× the builder before it matters.
max_run_cost and per-agent max_cost HALT the run when crossed. The crossing call is billed and recorded — then nothing further runs. Not an alert. A stop.
A chain parks on a pending approval until a human grants it — from any terminal. Silence past the timeout is a denial. Your agents never assume a yes.
Median, p90, and worst-case cost from your own run history — before the first token is bought. A warning fires when p90 clears your cap.
$ just sessions RUN PIPELINE PHASES COST STATUS 0139 feat/search-index 4/4 $2.13 ✔ complete 0140 fix/rate-limiter 3/4 $3.87 ⏸ awaiting approval (deploy) 0141 feat/checkout 4/4 $1.94 ✔ complete 0142 feat/checkout-v2 3/4 $4.71 ■ HALTED — max_run_cost $5.00 $ just estimate feat/checkout-v3 estimate from 12 prior runs: median $2.04 p90 $5.62 worst $7.90 ⚠ WARNING: p90 ($5.62) clears max_run_cost ($5.00) $
FAQ
No. SDKs observe from the sidelines — they can tell you what happened, not stop it. Flightbox is a managed runway: you run your agent pipelines through it. Deterministic Python owns the loop and agents are bounded nodes inside it, which is why a budget gate can actually halt a run mid-flight instead of just logging that it went over.
The call that crosses the cap is billed and recorded — you can't un-buy tokens already in flight. Then nothing further runs. The halt, the crossing call, and the full trace up to that point are all in the recording.
The chain parks on the pending approval and waits. You can grant it from any terminal. If the timeout passes in silence, that's a denial — the run does not proceed on an assumed yes.
Entirely on your machine: a SQLite trace inside your repo, read by a local UI. Nothing is sent anywhere — there is no account and no server involved. Flightbox Cloud will add an optional hosted store so teammates can share timelines across repos; until you opt into that, your traces never leave your disk.
A locked-in rate and a founding place in Flightbox Cloud — plus the knowledge that you funded it. You do not get extra features in the core, because the core has no locked features: budget caps, the halt, approval gates, cost estimates, replay and roll-ups are all in the free MIT release right now. If you want the software, take it for free. Pay only if you want the org-wide layer to exist and want your price fixed before it does.
It's the whole point, which is exactly why it shouldn't be behind a paywall. A budget cap you have to buy before it protects you is a hostage, not a safety feature — and a governor that only some runs have is worse than useless, because you stop checking. So the stop is free on every install, forever. What costs money is proving it: showing your client's compliance team, or your CFO, that the policy held across every repo and every developer, months later, in a form they'll accept. That's a different job, and it's the one Cloud does.
The dollars your agents actually spend on runs that Flightbox is governing — the same number the recorder already sums for you. Not your Flightbox bill, not your total cloud bill. Plans are banded by it rather than by seats because seats are meaningless here: the thing consuming the product is an agent, not a person. Go over your band and you're billed $12 per additional $1,000 governed, and I'll nudge you to move up after three months. Your governor is never switched off for a billing reason — halting your halt over an invoice would be indefensible.
It's in active development and I won't quote a date I can't stand behind. Your card isn't charged for 14 days, you can cancel any time from the receipt Stripe emails you, and if you're still waiting longer than you think is reasonable, email me and I'll refund you — no argument.
Nothing, and that's deliberate. Fork it, run it, modify it, ship it inside your own product — MIT means what it says. Cloud is worth paying for only if you'd rather not run a shared trace store, auth, and alerting yourself. That's the honest trade, and if self-hosting suits you better, take the free core with my blessing.
Cleared to operate
One command stamps it into any repo. Recording, replay and hard budget stops cost nothing and always will — no account, no key, no catch.