COST·FEED
Live model and agent cost feed: GPT-4o $0.031 up, claude $0.018 down, loop#42 $4,287 spiking, haiku $0.002 down, o3-mini $0.014 up, embed $0.0001 down, agent#4821 $612 up, tool.code $3.91 spiking, sonnet $0.022 up, rerank $0.008 down, loop#17 $1,204 spiking, gpt-4o-mini $0.006 down.
AI Capital
sign inget started
A

AGENT TASK TREE

⚠ $4,200 OVER BUDGET
run()$0.04
plan$0.31
synth$0.12
web$0.18
code·loop$4.12
model$0.22
eval$0.09
predicted cost / branchdepth=3 · 7 nodes
B

COST FORECAST

PROJ ▲ +0%
ceiling $4,200
cum. spend → projectedt+10 iterations
C

LINT OUTPUT

$ aicapital lint ./agent-loop.ts
⚠ line 42 — projected cost: $4,287
  └ recursion depth: unbounded
  └ model: claude (downgrade → haiku?)
  └ 18 tool calls / iteration
✔ suggested fix → $127
  ─────────────────────────────
  saved/run $4,160 · static · 63ms
aicapital@v0 — pre-flightexit 1 · 1 warning
D

BUDGET GUARD

96% · CRITICAL
96%of budget consumed
$4,032
spent
$4,200
cap
$168
left
auto-halt at 100%policy: hard-stop

See the bill before the agent runs.

Linter for agentic logic that predicts hidden costs — static analysis, before a single token is spent.

$ aicapital lint →free · no card · first lint in 60s
$2.4Mrunaway loops prevented
18,000budgets enforced
63msstatic analysis / run