Cost governance that never needs to see the content

Clinical documentation and intake agents run on text that cannot leave its boundary casually. That rules out a whole class of cost tooling: anything that works by capturing prompts and completions is adding a second store of exactly the material the rest of the stack is built to contain.

Try for free Read the docs

What a provider dashboard leaves you with

01
Observability tools that log request bodies put clinical text somewhere new, which is a governance question before it is a cost one.
02
Per-department attribution is impossible from one shared credential, so cost lands on IT rather than the service that incurred it.
03
A proxy or gateway in the call path becomes something else that has to be reviewed, and something else that can fail.

What Capsera does about it

Prompts and completions are never stored: Capsera measures cost and identity. It does not need the content of a call to attribute or to block one.
No proxy, no gateway: The SDK patches the provider libraries in-process at init. Nothing new sits in the network path between your application and the provider.
Fail-open by design: If Capsera is unavailable, your calls still go through. The governance layer is never the reason a clinical workflow stops.

The number this is judged on

Spend per department, per service line

None of this is assumed. We scanned 133 public agent repositories and found a cost defect in most of the ones making live model calls.

Read the method and limitations

import capsera
from anthropic import Anthropic
 
capsera.init(
    agent="reasoning_agent",
    run_id=ctx.run_id,
)
 
# checked before the call
capsera.budget(
    monthly=500,
    on_exceed="block",
)
 
with capsera.agent("summarizer"):
    client = Anthropic()
    resp = client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=1024,
    )
 
# no proxy, no gateway
assert resp.usage.cost < 0.02
 
if capsera.over_budget():
    raise BudgetExceeded()
 
@capsera.track()
def extract(doc):
    return chain(doc)
 
# cheaper route, same output
capsera.route(prefer="cost")

    on_exceed="block",
)
 
with capsera.agent("summarizer"):
    client = Anthropic()
    resp = client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=1024,
    )
 
# no proxy, no gateway
assert resp.usage.cost < 0.02
 
if capsera.over_budget():
    raise BudgetExceeded()
 
@capsera.track()
def extract(doc):
    return chain(doc)
 
# cheaper route, same output
capsera.route(prefer="cost")
 
spend = capsera.usage(
    group_by="agent",
    window="30d",
)
 
for a, cost in spend:
    print(a, f"${cost:.2f}")
 
# fails open, always
capsera.flush()

assert resp.usage.cost < 0.02
 
if capsera.over_budget():
    raise BudgetExceeded()
 
@capsera.track()
def extract(doc):
    return chain(doc)
 
# cheaper route, same output
capsera.route(prefer="cost")
 
spend = capsera.usage(
    group_by="agent",
    window="30d",
)
 
for a, cost in spend:
    print(a, f"${cost:.2f}")
 
# fails open, always
capsera.flush()
 
for task in queue:
    run(task)
 
log("run complete")
import capsera
from anthropic import Anthropic
 
capsera.init(
    agent="reasoning_agent",
    run_id=ctx.run_id,
)

 
spend = capsera.usage(
    group_by="agent",
    window="30d",
)
 
for a, cost in spend:
    print(a, f"${cost:.2f}")
 
# fails open, always
capsera.flush()
 
for task in queue:
    run(task)
 
log("run complete")
import capsera
from anthropic import Anthropic
 
capsera.init(
    agent="reasoning_agent",
    run_id=ctx.run_id,
)
 
# checked before the call
capsera.budget(
    monthly=500,
    on_exceed="block",
)
 
with capsera.agent("summarizer"):
    client = Anthropic()
    resp = client.messages.create(
        model="claude-sonnet-4-6",

 
for task in queue:
    run(task)
 
log("run complete")
import capsera
from anthropic import Anthropic
 
capsera.init(
    agent="reasoning_agent",
    run_id=ctx.run_id,
)
 
# checked before the call
capsera.budget(
    monthly=500,
    on_exceed="block",
)
 
with capsera.agent("summarizer"):
    client = Anthropic()
    resp = client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=1024,
    )
 
# no proxy, no gateway
assert resp.usage.cost < 0.02
 
if capsera.over_budget():
    raise BudgetExceeded()
 
@capsera.track()
def extract(doc):

)
 
# checked before the call
capsera.budget(
    monthly=500,
    on_exceed="block",
)
 
with capsera.agent("summarizer"):
    client = Anthropic()
    resp = client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=1024,
    )
 
# no proxy, no gateway
assert resp.usage.cost < 0.02
 
if capsera.over_budget():
    raise BudgetExceeded()
 
@capsera.track()
def extract(doc):
    return chain(doc)
 
# cheaper route, same output
capsera.route(prefer="cost")
 
spend = capsera.usage(
    group_by="agent",
    window="30d",
)
 
for a, cost in spend:

        model="claude-sonnet-4-6",
        max_tokens=1024,
    )
 
# no proxy, no gateway
assert resp.usage.cost < 0.02
 
if capsera.over_budget():
    raise BudgetExceeded()
 
@capsera.track()
def extract(doc):
    return chain(doc)
 
# cheaper route, same output
capsera.route(prefer="cost")
 
spend = capsera.usage(
    group_by="agent",
    window="30d",
)
 
for a, cost in spend:
    print(a, f"${cost:.2f}")
 
# fails open, always
capsera.flush()
 
for task in queue:
    run(task)
 
log("run complete")
import capsera
from anthropic import Anthropic

def extract(doc):
    return chain(doc)
 
# cheaper route, same output
capsera.route(prefer="cost")
 
spend = capsera.usage(
    group_by="agent",
    window="30d",
)
 
for a, cost in spend:
    print(a, f"${cost:.2f}")
 
# fails open, always
capsera.flush()
 
for task in queue:
    run(task)
 
log("run complete")
import capsera
from anthropic import Anthropic
 
capsera.init(
    agent="reasoning_agent",
    run_id=ctx.run_id,
)
 
# checked before the call
capsera.budget(
    monthly=500,
    on_exceed="block",
)

for a, cost in spend:
    print(a, f"${cost:.2f}")
 
# fails open, always
capsera.flush()
 
for task in queue:
    run(task)
 
log("run complete")
import capsera
from anthropic import Anthropic
 
capsera.init(
    agent="reasoning_agent",
    run_id=ctx.run_id,
)
 
# checked before the call
capsera.budget(
    monthly=500,
    on_exceed="block",
)
 
with capsera.agent("summarizer"):
    client = Anthropic()
    resp = client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=1024,
    )
 
# no proxy, no gateway
assert resp.usage.cost < 0.02

from anthropic import Anthropic
 
capsera.init(
    agent="reasoning_agent",
    run_id=ctx.run_id,
)
 
# checked before the call
capsera.budget(
    monthly=500,
    on_exceed="block",
)
 
with capsera.agent("summarizer"):
    client = Anthropic()
    resp = client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=1024,
    )
 
# no proxy, no gateway
assert resp.usage.cost < 0.02
 
if capsera.over_budget():
    raise BudgetExceeded()
 
@capsera.track()
def extract(doc):
    return chain(doc)
 
# cheaper route, same output
capsera.route(prefer="cost")
 
spend = capsera.usage(

)
 
with capsera.agent("summarizer"):
    client = Anthropic()
    resp = client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=1024,
    )
 
# no proxy, no gateway
assert resp.usage.cost < 0.02
 
if capsera.over_budget():
    raise BudgetExceeded()
 
@capsera.track()
def extract(doc):
    return chain(doc)
 
# cheaper route, same output
capsera.route(prefer="cost")
 
spend = capsera.usage(
    group_by="agent",
    window="30d",
)
 
for a, cost in spend:
    print(a, f"${cost:.2f}")
 
# fails open, always
capsera.flush()
 
for task in queue:

 
if capsera.over_budget():
    raise BudgetExceeded()
 
@capsera.track()
def extract(doc):
    return chain(doc)
 
# cheaper route, same output
capsera.route(prefer="cost")
 
spend = capsera.usage(
    group_by="agent",
    window="30d",
)
 
for a, cost in spend:
    print(a, f"${cost:.2f}")
 
# fails open, always
capsera.flush()
 
for task in queue:
    run(task)
 
log("run complete")
import capsera
from anthropic import Anthropic
 
capsera.init(
    agent="reasoning_agent",
    run_id=ctx.run_id,
)

import capsera
from anthropic import Anthropic
 
capsera.init(
    agent="reasoning_agent",
    run_id=ctx.run_id,
)
 
# checked before the call
capsera.budget(
    monthly=500,
    on_exceed="block",
)
 
with capsera.agent("summarizer"):
    client = Anthropic()
    resp = client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=1024,
    )
 
# no proxy, no gateway
assert resp.usage.cost < 0.02
 
if capsera.over_budget():
    raise BudgetExceeded()
 
@capsera.track()
def extract(doc):
    return chain(doc)
 
# cheaper route, same output
capsera.route(prefer="cost")

    on_exceed="block",
)
 
with capsera.agent("summarizer"):
    client = Anthropic()
    resp = client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=1024,
    )
 
# no proxy, no gateway
assert resp.usage.cost < 0.02
 
if capsera.over_budget():
    raise BudgetExceeded()
 
@capsera.track()
def extract(doc):
    return chain(doc)
 
# cheaper route, same output
capsera.route(prefer="cost")
 
spend = capsera.usage(
    group_by="agent",
    window="30d",
)
 
for a, cost in spend:
    print(a, f"${cost:.2f}")
 
# fails open, always
capsera.flush()

assert resp.usage.cost < 0.02
 
if capsera.over_budget():
    raise BudgetExceeded()
 
@capsera.track()
def extract(doc):
    return chain(doc)
 
# cheaper route, same output
capsera.route(prefer="cost")
 
spend = capsera.usage(
    group_by="agent",
    window="30d",
)
 
for a, cost in spend:
    print(a, f"${cost:.2f}")
 
# fails open, always
capsera.flush()
 
for task in queue:
    run(task)
 
log("run complete")
import capsera
from anthropic import Anthropic
 
capsera.init(
    agent="reasoning_agent",
    run_id=ctx.run_id,
)

 
spend = capsera.usage(
    group_by="agent",
    window="30d",
)
 
for a, cost in spend:
    print(a, f"${cost:.2f}")
 
# fails open, always
capsera.flush()
 
for task in queue:
    run(task)
 
log("run complete")
import capsera
from anthropic import Anthropic
 
capsera.init(
    agent="reasoning_agent",
    run_id=ctx.run_id,
)
 
# checked before the call
capsera.budget(
    monthly=500,
    on_exceed="block",
)
 
with capsera.agent("summarizer"):
    client = Anthropic()
    resp = client.messages.create(
        model="claude-sonnet-4-6",

 
for task in queue:
    run(task)
 
log("run complete")
import capsera
from anthropic import Anthropic
 
capsera.init(
    agent="reasoning_agent",
    run_id=ctx.run_id,
)
 
# checked before the call
capsera.budget(
    monthly=500,
    on_exceed="block",
)
 
with capsera.agent("summarizer"):
    client = Anthropic()
    resp = client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=1024,
    )
 
# no proxy, no gateway
assert resp.usage.cost < 0.02
 
if capsera.over_budget():
    raise BudgetExceeded()
 
@capsera.track()
def extract(doc):

)
 
# checked before the call
capsera.budget(
    monthly=500,
    on_exceed="block",
)
 
with capsera.agent("summarizer"):
    client = Anthropic()
    resp = client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=1024,
    )
 
# no proxy, no gateway
assert resp.usage.cost < 0.02
 
if capsera.over_budget():
    raise BudgetExceeded()
 
@capsera.track()
def extract(doc):
    return chain(doc)
 
# cheaper route, same output
capsera.route(prefer="cost")
 
spend = capsera.usage(
    group_by="agent",
    window="30d",
)
 
for a, cost in spend:

        model="claude-sonnet-4-6",
        max_tokens=1024,
    )
 
# no proxy, no gateway
assert resp.usage.cost < 0.02
 
if capsera.over_budget():
    raise BudgetExceeded()
 
@capsera.track()
def extract(doc):
    return chain(doc)
 
# cheaper route, same output
capsera.route(prefer="cost")
 
spend = capsera.usage(
    group_by="agent",
    window="30d",
)
 
for a, cost in spend:
    print(a, f"${cost:.2f}")
 
# fails open, always
capsera.flush()
 
for task in queue:
    run(task)
 
log("run complete")
import capsera
from anthropic import Anthropic

def extract(doc):
    return chain(doc)
 
# cheaper route, same output
capsera.route(prefer="cost")
 
spend = capsera.usage(
    group_by="agent",
    window="30d",
)
 
for a, cost in spend:
    print(a, f"${cost:.2f}")
 
# fails open, always
capsera.flush()
 
for task in queue:
    run(task)
 
log("run complete")
import capsera
from anthropic import Anthropic
 
capsera.init(
    agent="reasoning_agent",
    run_id=ctx.run_id,
)
 
# checked before the call
capsera.budget(
    monthly=500,
    on_exceed="block",
)

for a, cost in spend:
    print(a, f"${cost:.2f}")
 
# fails open, always
capsera.flush()
 
for task in queue:
    run(task)
 
log("run complete")
import capsera
from anthropic import Anthropic
 
capsera.init(
    agent="reasoning_agent",
    run_id=ctx.run_id,
)
 
# checked before the call
capsera.budget(
    monthly=500,
    on_exceed="block",
)
 
with capsera.agent("summarizer"):
    client = Anthropic()
    resp = client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=1024,
    )
 
# no proxy, no gateway
assert resp.usage.cost < 0.02

from anthropic import Anthropic
 
capsera.init(
    agent="reasoning_agent",
    run_id=ctx.run_id,
)
 
# checked before the call
capsera.budget(
    monthly=500,
    on_exceed="block",
)
 
with capsera.agent("summarizer"):
    client = Anthropic()
    resp = client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=1024,
    )
 
# no proxy, no gateway
assert resp.usage.cost < 0.02
 
if capsera.over_budget():
    raise BudgetExceeded()
 
@capsera.track()
def extract(doc):
    return chain(doc)
 
# cheaper route, same output
capsera.route(prefer="cost")
 
spend = capsera.usage(

)
 
with capsera.agent("summarizer"):
    client = Anthropic()
    resp = client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=1024,
    )
 
# no proxy, no gateway
assert resp.usage.cost < 0.02
 
if capsera.over_budget():
    raise BudgetExceeded()
 
@capsera.track()
def extract(doc):
    return chain(doc)
 
# cheaper route, same output
capsera.route(prefer="cost")
 
spend = capsera.usage(
    group_by="agent",
    window="30d",
)
 
for a, cost in spend:
    print(a, f"${cost:.2f}")
 
# fails open, always
capsera.flush()
 
for task in queue:

 
if capsera.over_budget():
    raise BudgetExceeded()
 
@capsera.track()
def extract(doc):
    return chain(doc)
 
# cheaper route, same output
capsera.route(prefer="cost")
 
spend = capsera.usage(
    group_by="agent",
    window="30d",
)
 
for a, cost in spend:
    print(a, f"${cost:.2f}")
 
# fails open, always
capsera.flush()
 
for task in queue:
    run(task)
 
log("run complete")
import capsera
from anthropic import Anthropic
 
capsera.init(
    agent="reasoning_agent",
    run_id=ctx.run_id,
)

import capsera
from anthropic import Anthropic
 
capsera.init(
    agent="reasoning_agent",
    run_id=ctx.run_id,
)
 
# checked before the call
capsera.budget(
    monthly=500,
    on_exceed="block",
)
 
with capsera.agent("summarizer"):
    client = Anthropic()
    resp = client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=1024,
    )
 
# no proxy, no gateway
assert resp.usage.cost < 0.02
 
if capsera.over_budget():
    raise BudgetExceeded()
 
@capsera.track()
def extract(doc):
    return chain(doc)
 
# cheaper route, same output
capsera.route(prefer="cost")

    on_exceed="block",
)
 
with capsera.agent("summarizer"):
    client = Anthropic()
    resp = client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=1024,
    )
 
# no proxy, no gateway
assert resp.usage.cost < 0.02
 
if capsera.over_budget():
    raise BudgetExceeded()
 
@capsera.track()
def extract(doc):
    return chain(doc)
 
# cheaper route, same output
capsera.route(prefer="cost")
 
spend = capsera.usage(
    group_by="agent",
    window="30d",
)
 
for a, cost in spend:
    print(a, f"${cost:.2f}")
 
# fails open, always
capsera.flush()

assert resp.usage.cost < 0.02
 
if capsera.over_budget():
    raise BudgetExceeded()
 
@capsera.track()
def extract(doc):
    return chain(doc)
 
# cheaper route, same output
capsera.route(prefer="cost")
 
spend = capsera.usage(
    group_by="agent",
    window="30d",
)
 
for a, cost in spend:
    print(a, f"${cost:.2f}")
 
# fails open, always
capsera.flush()
 
for task in queue:
    run(task)
 
log("run complete")
import capsera
from anthropic import Anthropic
 
capsera.init(
    agent="reasoning_agent",
    run_id=ctx.run_id,
)

 
spend = capsera.usage(
    group_by="agent",
    window="30d",
)
 
for a, cost in spend:
    print(a, f"${cost:.2f}")
 
# fails open, always
capsera.flush()
 
for task in queue:
    run(task)
 
log("run complete")
import capsera
from anthropic import Anthropic
 
capsera.init(
    agent="reasoning_agent",
    run_id=ctx.run_id,
)
 
# checked before the call
capsera.budget(
    monthly=500,
    on_exceed="block",
)
 
with capsera.agent("summarizer"):
    client = Anthropic()
    resp = client.messages.create(
        model="claude-sonnet-4-6",

 
for task in queue:
    run(task)
 
log("run complete")
import capsera
from anthropic import Anthropic
 
capsera.init(
    agent="reasoning_agent",
    run_id=ctx.run_id,
)
 
# checked before the call
capsera.budget(
    monthly=500,
    on_exceed="block",
)
 
with capsera.agent("summarizer"):
    client = Anthropic()
    resp = client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=1024,
    )
 
# no proxy, no gateway
assert resp.usage.cost < 0.02
 
if capsera.over_budget():
    raise BudgetExceeded()
 
@capsera.track()
def extract(doc):

)
 
# checked before the call
capsera.budget(
    monthly=500,
    on_exceed="block",
)
 
with capsera.agent("summarizer"):
    client = Anthropic()
    resp = client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=1024,
    )
 
# no proxy, no gateway
assert resp.usage.cost < 0.02
 
if capsera.over_budget():
    raise BudgetExceeded()
 
@capsera.track()
def extract(doc):
    return chain(doc)
 
# cheaper route, same output
capsera.route(prefer="cost")
 
spend = capsera.usage(
    group_by="agent",
    window="30d",
)
 
for a, cost in spend:

        model="claude-sonnet-4-6",
        max_tokens=1024,
    )
 
# no proxy, no gateway
assert resp.usage.cost < 0.02
 
if capsera.over_budget():
    raise BudgetExceeded()
 
@capsera.track()
def extract(doc):
    return chain(doc)
 
# cheaper route, same output
capsera.route(prefer="cost")
 
spend = capsera.usage(
    group_by="agent",
    window="30d",
)
 
for a, cost in spend:
    print(a, f"${cost:.2f}")
 
# fails open, always
capsera.flush()
 
for task in queue:
    run(task)
 
log("run complete")
import capsera
from anthropic import Anthropic

def extract(doc):
    return chain(doc)
 
# cheaper route, same output
capsera.route(prefer="cost")
 
spend = capsera.usage(
    group_by="agent",
    window="30d",
)
 
for a, cost in spend:
    print(a, f"${cost:.2f}")
 
# fails open, always
capsera.flush()
 
for task in queue:
    run(task)
 
log("run complete")
import capsera
from anthropic import Anthropic
 
capsera.init(
    agent="reasoning_agent",
    run_id=ctx.run_id,
)
 
# checked before the call
capsera.budget(
    monthly=500,
    on_exceed="block",
)

for a, cost in spend:
    print(a, f"${cost:.2f}")
 
# fails open, always
capsera.flush()
 
for task in queue:
    run(task)
 
log("run complete")
import capsera
from anthropic import Anthropic
 
capsera.init(
    agent="reasoning_agent",
    run_id=ctx.run_id,
)
 
# checked before the call
capsera.budget(
    monthly=500,
    on_exceed="block",
)
 
with capsera.agent("summarizer"):
    client = Anthropic()
    resp = client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=1024,
    )
 
# no proxy, no gateway
assert resp.usage.cost < 0.02

from anthropic import Anthropic
 
capsera.init(
    agent="reasoning_agent",
    run_id=ctx.run_id,
)
 
# checked before the call
capsera.budget(
    monthly=500,
    on_exceed="block",
)
 
with capsera.agent("summarizer"):
    client = Anthropic()
    resp = client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=1024,
    )
 
# no proxy, no gateway
assert resp.usage.cost < 0.02
 
if capsera.over_budget():
    raise BudgetExceeded()
 
@capsera.track()
def extract(doc):
    return chain(doc)
 
# cheaper route, same output
capsera.route(prefer="cost")
 
spend = capsera.usage(

)
 
with capsera.agent("summarizer"):
    client = Anthropic()
    resp = client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=1024,
    )
 
# no proxy, no gateway
assert resp.usage.cost < 0.02
 
if capsera.over_budget():
    raise BudgetExceeded()
 
@capsera.track()
def extract(doc):
    return chain(doc)
 
# cheaper route, same output
capsera.route(prefer="cost")
 
spend = capsera.usage(
    group_by="agent",
    window="30d",
)
 
for a, cost in spend:
    print(a, f"${cost:.2f}")
 
# fails open, always
capsera.flush()
 
for task in queue:

 
if capsera.over_budget():
    raise BudgetExceeded()
 
@capsera.track()
def extract(doc):
    return chain(doc)
 
# cheaper route, same output
capsera.route(prefer="cost")
 
spend = capsera.usage(
    group_by="agent",
    window="30d",
)
 
for a, cost in spend:
    print(a, f"${cost:.2f}")
 
# fails open, always
capsera.flush()
 
for task in queue:
    run(task)
 
log("run complete")
import capsera
from anthropic import Anthropic
 
capsera.init(
    agent="reasoning_agent",
    run_id=ctx.run_id,
)

Govern the spend without touching the record.

Stop runaway agents before they become runaway invoices.

Try for free