Caveman
Caveman

Cut 65% of your tokens.

74kon GitHub#1 on Hacker News

≈35% kept65% cut

Starred by engineers at

01Open toolkit

Caveman for developers.

Local tools for agents that say less, remember what matters, and keep every technical detail intact. Each product below is working—not a poster.

01available

Caveman

The skill, and the engine.

Compression workbench local demo72.8k
intensity
156 est. tok
Output · caveman49 est. tok

Auth: validate email pre-insert; reject disposable domains; keep `oauth_callback`; log `organization_id` + `request_id`; omit query/secrets; test valid/malformed/duplicate; return `invalid_email`.

illustrative rewrite · not production engine output

Token map · first 48 kept removed code
01voiceactive
02structuralavailable
03CCRunused
04recoverynot needed
02available

Caveman Code

The flagship CLI.

caveman-code live
$caveman-code "ship /healthz + test"
4 compression layers active
router.go · router_test.go
go test ./... · passed
4 layers active
03soon

caveman wrap

Byte-safe local proxy.

agentbytes in
caverecord
modelbytes out
request
unchanged
cache hint
upstream only
price
catalog
savings
not claimed
04available

Cavemem

Memory offload.

auth decision2 hits
JWT org from auth context0.94
record = pass-through0.89
05available

Cavekit

Spec-driven dev.

SPEC.mddurable context

loop

/spec

SPEC.md

/build

2 files

/check

tests

# auth invariant

org_id := jwt.claim

never request.body

failure → spec invariant
06available

Caveman Mode

Browser extension.

chatcaveman on
Why is build failing? Please explain what I need to change.
Missing tsconfig.paths. Add alias. Rebuild.
07available

SDKs & MCP

TypeScript and Python.

TypeScript@caveman/sdk

const cave = new Cave({ baseURL })

await cave.compress(json)

return recoveryHandle

same gateway contract
Pythoncaveman_cloud

cave = Cave(base_url=url)

result = cave.compress(data)

print(recovery_handle)

same gateway contract
MCPcaveman-mcp

tool caveman_compress

input JSON · logs · schemas

output inferred · CCR handle

same gateway contract
02Managed control plane

Caveman for enterprise.

Measure provider usage, rank inferred headroom, and run guarded experiments. Unknown usage stays unavailable, unknown prices stay unpriced, and automatic signing remains off.

acme-prod · requestsdemo data
Caveman Cloud requests
Truthful catalog subtotals

Complete provider-reported usage is priced from the public catalog. Incomplete usage and unknown models remain visibly unpriced, never guessed.

03Models & research

Caveman for research.

Compression moved from prompt into weights, then tested in public. Open models, reproducible numbers, and experiments that say where they lose.

Caveman Labs · Research division4 papers

CaveGemma

fine-tune of google/gemma-4-31B-it · QLoRA r16 · MIT · weights inherit Gemma terms

Get the weights →
27%
fewer output tokens · 193 pairs
96–100%
code-fence exactness
0.91–0.98
semantic cosine
534MB
LoRA adapter
Specimen · logs as pixelsFig. 00
Input · agent system promptOutput · −65%

The proof is public.

Every number here is live and cited — held to the same honesty we hold our own metering to.

Fewer words. Same work.

The open stack is public today — read the source, send a patch. The managed Cloud is in private development; leave your email and we'll reach out the moment a spot opens.

No spam · one email when your invite is ready