Metadata-Version: 2.4
Name: cacheeconomics
Version: 0.2.0
Summary: Prompt-cache cost analysis: a dated provider registry, a cost model that separates the four token classes, and an analyzer that turns a trace into a report you can forward.
Author-email: Tanisha Katara <tanisha.bk@gmail.com>
License: 
                                         Apache License
                                   Version 2.0, January 2004
                                http://www.apache.org/licenses/
        
           TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
        
           1. Definitions.
        
              "License" shall mean the terms and conditions for use, reproduction,
              and distribution as defined by Sections 1 through 9 of this document.
        
              "Licensor" shall mean the copyright owner or entity authorized by
              the copyright owner that is granting the License.
        
              "Legal Entity" shall mean the union of the acting entity and all
              other entities that control, are controlled by, or are under common
              control with that entity. For the purposes of this definition,
              "control" means (i) the power, direct or indirect, to cause the
              direction or management of such entity, whether by contract or
              otherwise, or (ii) ownership of fifty percent (50%) or more of the
              outstanding shares, or (iii) beneficial ownership of such entity.
        
              "You" (or "Your") shall mean an individual or Legal Entity
              exercising permissions granted by this License.
        
              "Source" form shall mean the preferred form for making modifications,
              including but not limited to software source code, documentation
              source, and configuration files.
        
              "Object" form shall mean any form resulting from mechanical
              transformation or translation of a Source form, including but
              not limited to compiled object code, generated documentation,
              and conversions to other media types.
        
              "Work" shall mean the work of authorship, whether in Source or
              Object form, made available under the License, as indicated by a
              copyright notice that is included in or attached to the work
              (an example is provided in the Appendix below).
        
              "Derivative Works" shall mean any work, whether in Source or Object
              form, that is based on (or derived from) the Work and for which the
              editorial revisions, annotations, elaborations, or other modifications
              represent, as a whole, an original work of authorship. For the purposes
              of this License, Derivative Works shall not include works that remain
              separable from, or merely link (or bind by name) to the interfaces of,
              the Work and Derivative Works thereof.
        
              "Contribution" shall mean any work of authorship, including
              the original version of the Work and any modifications or additions
              to that Work or Derivative Works thereof, that is intentionally
              submitted to Licensor for inclusion in the Work by the copyright owner
              or by an individual or Legal Entity authorized to submit on behalf of
              the copyright owner. For the purposes of this definition, "submitted"
              means any form of electronic, verbal, or written communication sent
              to the Licensor or its representatives, including but not limited to
              communication on electronic mailing lists, source code control systems,
              and issue tracking systems that are managed by, or on behalf of, the
              Licensor for the purpose of discussing and improving the Work, but
              excluding communication that is conspicuously marked or otherwise
              designated in writing by the copyright owner as "Not a Contribution."
        
              "Contributor" shall mean Licensor and any individual or Legal Entity
              on behalf of whom a Contribution has been received by Licensor and
              subsequently incorporated within the Work.
        
           2. Grant of Copyright License. Subject to the terms and conditions of
              this License, each Contributor hereby grants to You a perpetual,
              worldwide, non-exclusive, no-charge, royalty-free, irrevocable
              copyright license to reproduce, prepare Derivative Works of,
              publicly display, publicly perform, sublicense, and distribute the
              Work and such Derivative Works in Source or Object form.
        
           3. Grant of Patent License. Subject to the terms and conditions of
              this License, each Contributor hereby grants to You a perpetual,
              worldwide, non-exclusive, no-charge, royalty-free, irrevocable
              (except as stated in this section) patent license to make, have made,
              use, offer to sell, sell, import, and otherwise transfer the Work,
              where such license applies only to those patent claims licensable
              by such Contributor that are necessarily infringed by their
              Contribution(s) alone or by combination of their Contribution(s)
              with the Work to which such Contribution(s) was submitted. If You
              institute patent litigation against any entity (including a
              cross-claim or counterclaim in a lawsuit) alleging that the Work
              or a Contribution incorporated within the Work constitutes direct
              or contributory patent infringement, then any patent licenses
              granted to You under this License for that Work shall terminate
              as of the date such litigation is filed.
        
           4. Redistribution. You may reproduce and distribute copies of the
              Work or Derivative Works thereof in any medium, with or without
              modifications, and in Source or Object form, provided that You
              meet the following conditions:
        
              (a) You must give any other recipients of the Work or
                  Derivative Works a copy of this License; and
        
              (b) You must cause any modified files to carry prominent notices
                  stating that You changed the files; and
        
              (c) You must retain, in the Source form of any Derivative Works
                  that You distribute, all copyright, patent, trademark, and
                  attribution notices from the Source form of the Work,
                  excluding those notices that do not pertain to any part of
                  the Derivative Works; and
        
              (d) If the Work includes a "NOTICE" text file as part of its
                  distribution, then any Derivative Works that You distribute must
                  include a readable copy of the attribution notices contained
                  within such NOTICE file, excluding those notices that do not
                  pertain to any part of the Derivative Works, in at least one
                  of the following places: within a NOTICE text file distributed
                  as part of the Derivative Works; within the Source form or
                  documentation, if provided along with the Derivative Works; or,
                  within a display generated by the Derivative Works, if and
                  wherever such third-party notices normally appear. The contents
                  of the NOTICE file are for informational purposes only and
                  do not modify the License. You may add Your own attribution
                  notices within Derivative Works that You distribute, alongside
                  or as an addendum to the NOTICE text from the Work, provided
                  that such additional attribution notices cannot be construed
                  as modifying the License.
        
              You may add Your own copyright statement to Your modifications and
              may provide additional or different license terms and conditions
              for use, reproduction, or distribution of Your modifications, or
              for any such Derivative Works as a whole, provided Your use,
              reproduction, and distribution of the Work otherwise complies with
              the conditions stated in this License.
        
           5. Submission of Contributions. Unless You explicitly state otherwise,
              any Contribution intentionally submitted for inclusion in the Work
              by You to the Licensor shall be under the terms and conditions of
              this License, without any additional terms or conditions.
              Notwithstanding the above, nothing herein shall supersede or modify
              the terms of any separate license agreement you may have executed
              with Licensor regarding such Contributions.
        
           6. Trademarks. This License does not grant permission to use the trade
              names, trademarks, service marks, or product names of the Licensor,
              except as required for reasonable and customary use in describing the
              origin of the Work and reproducing the content of the NOTICE file.
        
           7. Disclaimer of Warranty. Unless required by applicable law or
              agreed to in writing, Licensor provides the Work (and each
              Contributor provides its Contributions) on an "AS IS" BASIS,
              WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
              implied, including, without limitation, any warranties or conditions
              of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
              PARTICULAR PURPOSE. You are solely responsible for determining the
              appropriateness of using or redistributing the Work and assume any
              risks associated with Your exercise of permissions under this License.
        
           8. Limitation of Liability. In no event and under no legal theory,
              whether in tort (including negligence), contract, or otherwise,
              unless required by applicable law (such as deliberate and grossly
              negligent acts) or agreed to in writing, shall any Contributor be
              liable to You for damages, including any direct, indirect, special,
              incidental, or consequential damages of any character arising as a
              result of this License or out of the use or inability to use the
              Work (including but not limited to damages for loss of goodwill,
              work stoppage, computer failure or malfunction, or any and all
              other commercial damages or losses), even if such Contributor
              has been advised of the possibility of such damages.
        
           9. Accepting Warranty or Additional Liability. While redistributing
              the Work or Derivative Works thereof, You may choose to offer,
              and charge a fee for, acceptance of support, warranty, indemnity,
              or other liability obligations and/or rights consistent with this
              License. However, in accepting such obligations, You may act only
              on Your own behalf and on Your sole responsibility, not on behalf
              of any other Contributor, and only if You agree to indemnify,
              defend, and hold each Contributor harmless for any liability
              incurred by, or claims asserted against, such Contributor by reason
              of your accepting any such warranty or additional liability.
        
           END OF TERMS AND CONDITIONS
        
           APPENDIX: How to apply the Apache License to your work.
        
              To apply the Apache License to your work, attach the following
              boilerplate notice, with the fields enclosed by brackets "[]"
              replaced with your own identifying information. (Don't include
              the brackets!)  The text should be enclosed in the appropriate
              comment syntax for the file format. We also recommend that a
              file or class name and description of purpose be included on the
              same "printed page" as the copyright notice for easier
              identification within third-party archives.
        
           Copyright [yyyy] [name of copyright owner]
        
           Licensed under the Apache License, Version 2.0 (the "License");
           you may not use this file except in compliance with the License.
           You may obtain a copy of the License at
        
               http://www.apache.org/licenses/LICENSE-2.0
        
           Unless required by applicable law or agreed to in writing, software
           distributed under the License is distributed on an "AS IS" BASIS,
           WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
           See the License for the specific language governing permissions and
           limitations under the License.
        
Keywords: prompt-caching,llm,cost,anthropic,finops,litellm
Classifier: Development Status :: 4 - Beta
Classifier: Intended Audience :: Developers
Classifier: Intended Audience :: System Administrators
Classifier: License :: OSI Approved :: Apache Software License
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3 :: Only
Classifier: Topic :: Software Development :: Libraries :: Python Modules
Classifier: Topic :: System :: Monitoring
Requires-Python: >=3.9
Description-Content-Type: text/markdown
License-File: LICENSE
License-File: NOTICE
Provides-Extra: litellm
Requires-Dist: litellm>=1.0; extra == "litellm"
Dynamic: license-file

# cacheeconomics

You point it at logs you already have. It tells you what your prompt caching
actually cost, which part of the prompt is wasting the money, and how much of
that is recoverable.

It runs on your machine, opens no sockets, and will not print a dollar figure it
cannot tie to an invoice.

## Why this is not just adding up tokens

Your provider bills input tokens at four different rates: fresh input at 1x, a
cache read at 0.1x, a five-minute cache write at 1.25x, a one-hour write at 2x.
Your usage counters report the totals. They do not report which decisions put
tokens in the expensive buckets, and that is where the money is.

Three specific ways it leaks:

A one-hour cache is only cheaper inside a band. It costs 2x to write where the
short one costs 1.25x, so it wins only when your requests are more than five
minutes and less than an hour apart. Faster than that, the cheap entry was still
alive and you overpaid. Slower, both had expired and you overpaid. We measured
this on live API calls: $0.190 for the 5m arm against $0.114 for the 1h arm on
the same prefix, 40.2% cheaper, with all eight probes matching predictions
written down before the run. The per-call receipts are kept with the source; ask and I will send them.

Below a model's minimum, a cache marker does nothing and nobody tells you. The
threshold is 512, 1024, 2048 or 4096 tokens depending on the model, and it does
not track model generation the way you would expect: `claude-opus-5` is 512,
`claude-opus-4-6` is 4096. Under it the provider quietly processes your request
uncached and reports `cache_creation_input_tokens: 0`. No error.

A write nobody reads is pure premium. You paid 1.25x or 2x for a prefix and then
changed something at the top of it. Usage counters cannot tell you that happened,
because "we rebuilt this instead of extending it" is a fact about prompt
structure, not about token totals.

## What you need to run it

One of these, in rough order of how little work they are for you:

Your Claude Code transcripts. Already on your disk, nothing to instrument. Run
`cacheeconomics claude-code`.

Your LiteLLM proxy logs. If you run LiteLLM you are already writing
`StandardLoggingPayload` records. That file is the input. This path needs a
recent enough LiteLLM to include `prompt_tokens_details`; without the cache
class split the tool will tell you it cannot read the rows rather than guess.

Request bodies your gateway already logs. Langfuse, Helicone and the LiteLLM
proxy all keep full bodies. This is the first path that can answer "what would a
different prompt layout have cost", because it can see prompt structure.

An instrumented capture, using the recorder in this package. Needs a small code
change on your side and gives the strongest answers.

The file you have decides what the tool will claim. It works this out from the
contents, not from what you tell it, so an export cannot ask for more confidence
than it earns.

## Install and run

```bash
pip install cacheeconomics
```

Python 3.9 or newer and nothing else. `pip install "cacheeconomics[litellm]"`
only if you are mounting the live proxy plugin.

The source repository is private. If you need to read the code or file
something, contact the author address on this package.

```bash
# your own Claude Code usage, read locally
cacheeconomics claude-code --allow-unreconciled

# a trace, reconciled against the bill you actually received
cacheeconomics analyze trace.jsonl --invoice-usd 4820.16

# LiteLLM proxy logs, unchanged
cacheeconomics analyze standard_logging.jsonl --from litellm --invoice-usd 4820.16

# Bedrock or Vertex: the rate comes off your cloud bill, not from us
cacheeconomics analyze trace.jsonl --target-id amazon-bedrock/converse --effective-rate 1.10

# compare cache placement policies over the same requests
cacheeconomics bakeoff trace.jsonl --by-agent

# check a cache config before you ship it
cacheeconomics checks --prefix-tokens 900 --model claude-opus-5 --breakpoints 5

# what does this build actually know?
cacheeconomics registry
```

`checks` works as a CI gate. Exit 0 means everything passed, 2 means something
failed, and 3 means nothing failed but something could not be evaluated, usually
a model with no dated registry entry. That third code exists because "I did not
check" is not "this passed", and collapsing them gives you a green build where
the check that catches silently-ignored cache markers never ran. Exit 1 means the
tool itself broke.

## What it refuses to tell you

This part matters more than the analysis, because it is what makes the analysis
worth reading.

No invoice, no dollars. Every figure stays hidden until it reconciles against
money that actually left your account, within 5%. Without `--invoice-usd` the
report opens with FIGURES WITHHELD and prints `[figure withheld]` where each
number would be. `--allow-unreconciled` releases them for internal drafts, and it
is a flag you have to type rather than a default, so publishing an unreconciled
number is always somebody's decision.

Structure and usage have to agree. Spend comes from usage counters; structural
findings come from segment sizes. If those two descriptions of the same prompt
disagree by more than 5%, you still get the findings and you do not get their
dollar figures.

Coverage is measured in money, not rows. A structural claim needs to cover 90% of
your billed tokens before a figure attaches to it. Nine small structured requests
next to one huge unstructured one is 90% of rows and can be 8% of the bill, and a
recommendation costed from 8% of your spend does not describe your workload.

Bedrock and Vertex are not priced here. AWS and Google operate and invoice those,
so the rates are not ours to publish and we do not borrow Anthropic's. Pass
`--effective-rate` from your own cloud bill, which is the better number anyway:
partner rates vary by region and endpoint type, and nobody at your volume pays
list.

Every number says what kind of number it is. Measured means observed in usage
fields. Modeled means projected, always as a range, with the pessimistic end as
the headline. Verified means observed in production after a change shipped.
Nothing gets promoted between them quietly.

## Privacy

Everything runs locally. Nothing in the package opens a socket, the registry is
packaged data, and every input is a path on your disk.

Prompt text is optional. Hashes, structure and token counts are enough for every
finding here.

Segment identifiers are keyed HMAC-SHA-256 rather than bare digests, because a
bare digest of a short policy line is confirmable by anyone holding a guess.
There is deliberately no `--key` flag: `argv` shows up in `ps` and lands in shell
history, so the key comes from `CACHEECONOMICS_HMAC_KEY` or a `--key-file` whose
permissions get checked before it is read.

Keying and scoping are different things, and the difference matters on a shared
gateway. A key stops someone guessing your content. It does nothing about
equality, because equality is the whole job of an identifier: under one key,
identical text produces an identical id no matter who sent it. So identity is
scoped to the tenant as well. Pass `--tenant` on a multi-tenant export and two
tenants sending the same policy block stop sharing an id. That is also the
arithmetically correct answer, since caches are isolated per tenant and those two
requests could never have shared a cache entry.

## What is in here

```
harness/cacheeconomics/
  registry.py   dated capability and pricing registry, refuses to guess
  cost.py       the four token classes, date-effective pricing
  money.py      a dollar figure that cannot be printed until it is released
  trace.py      ingest, tiering, reconciliation
  analyzer.py   the finding rules (EFF-1, VOL-1, MIN-1, TTL-1, FAN-1, ...)
  report.py     text and self-contained HTML
  simulate.py   replay a trace against a modelled cache; the policy bake-off
  allocate.py   placement policies, including a model of LiteLLM auto-injection
  tiers.py      the DP tier allocator and an exact evaluator that agrees with it
  monitor.py    runtime diagnostics, bounded state
  plugin.py     a gateway plugin that decides inside the request path
  recorder.py   instrumented capture
  segment.py    the single traversal of a request body
  cli.py        the commands above
  data/         providers.json, pricing.json
```

The measurement work behind the numbers above lives in the source repository
rather than in the installed package: the live TTL experiment and its per-call
receipts, the landscape review with provenance pinned by commit and SHA-256, and
the findings filed against OpenHands, SWE-agent and browser-use. Ask if you want
them.

Eight provider surfaces are described. The recorded rates are Anthropic
first-party list prices, and the table names which surfaces they are valid for:
`anthropic/direct` and `anthropic/claude-platform-on-aws`. A test fails if a
model is ever added there without a dated rate.

The scope is default-deny, so a surface earns those rates by being named rather
than by being absent from an exclusion list. The alternative is how this went
wrong once already: before the scope existed, a Bedrock trace priced straight off
the Anthropic table and produced a confident total no AWS bill would match.

## Running it live

`plugin.py` can sit inside a LiteLLM proxy and place cache markers on outgoing
requests. It observes by default and only mutates when you ask.

The observing half is tested here. The mutating half is not: nothing in this
repository has watched a real LiteLLM proxy forward one of these requests, so we
do not know whether `cache_control` on a message content block survives LiteLLM's
translation to the Anthropic wire format. If it gets stripped you would get churn
and no cache writes while the plugin's own counters cheerfully reported
placements that never arrived. That is why `mutate` defaults to False and why
this paragraph exists.

## Testing

1011 tests, no dependencies beyond pytest, run in the source repository. CI runs
them on Python 3.9 and 3.13 with only pytest installed, so an accidental
dependency fails the build rather than shipping to you.

## License

Apache-2.0. Copyright 2026 KCG Consulting LLC.
