Send less repeated context.
Use prompt compression and compaction to reduce context overhead. Inspect token usage and savings.
CI for teams shipping with AI
Less AI spend.
Faster CI.
Your agents write more code. Compress helps CI reuse compatible checks, run changed work, and show exactly what executed. LLM optimization is included.
Connect your application's LLM requests, your CI workflows, or both.
No monthly fee · CI from $0.003/min at two vCPUs
$3 introductory compute credit, valid for 7 days. Provider LLM usage is separate.
The Compress optimization layerHover to reveal reuseTouch to explore
Context shrinksVerified work is reusedChanged checks run
Same work. Less compute.Illustrative flow
LLM requestsLess contextFresh checksExecutedCached resultsReused
Build more.Repeat less.
One idea. Two places to save.
Don't pay twice
for the same work.
Repeated context costs tokens. Repeated verification costs time. Compress brings LLM efficiency and CI reuse into one workspace.
Checks in the next CI runResult
//app:coreReused//app:apiReused//app:changedExecutedReuse requires matching inputs, tools, and platform.
CI optimization
An agent checks it.
CI can reuse it.
An agent verifies its change. The pipeline starts. With Bazel and Compress's private cache, compatible build and test actions can reuse the work that already passed.
Changed inputs still run. Required checks still apply. Your team gets a clear record of what executed and what was reused.
Set up Bazel with CompressUse Bazel. Share the results.For developers and coding agents
- 01
Pin your Bazel version.
Use your repository’s checked-in build and test targets. Bazelisk can download the right version for local runs.
- 02
Connect your private cache.
For enrolled repositories, use the authorized Compress cache in the same qualified runtime as CI. Cache misses still execute.
- 03
Bring your agents along.
Copy repository guidance from Actions settings → repository → Agent access into AGENTS.md or your agent’s project instructions.
Bazel handles the dependency graph and cache keys. Compress provides the private cache, access controls, and visibility into your CI runs.
LLM optimization
LLM optimization.
Included.
Long conversations and repeated context add up. Reduce that overhead before inference, then route requests within the model policy you choose.
Connect through the inference API to combine prompt compression, model routing, and cost controls. See the tokens and costs behind the requests you send through Compress.
Connect the inference APIPrompt contextCompressLeaner prompt
Request continues to your model
Prompt compression and compaction, before inference.
Choose models within your policy.
Set approved providers and a quality threshold. Compress routes requests to a qualifying available model.
Know where your budget goes.
Attribute integrated model usage to teams, customers, and features. Track spend and set workspace budgets.
No monthly platform fee
Optimization included.
Pay for CI you run.
Start with LLM optimization, one CI workflow, or both. One workspace. Usage-based compute.
Included workspace
Make your AI budget go further.
$0monthly platform fee
Connect your provider keys. Your provider bills model usage directly.
Start optimizing- Prompt compression, model routing, and spend visibility
- Optimize up to $1,000/month in routed provider spend
- Up to 5 million tokens saved each month
- Invite up to 5 teammates
- Workspace budgets and alerts
Pay-as-you-go CI
Fewer repeats. A smaller build bill.
$0.09USD / allocated vCPU-hour
$0.003/minute at two allocated vCPUs. Metered by the second.
Bring your first workflow- No monthly minimum or per-seat compute fee
- Import supported Linux workflow YAML from GitHub
- Reuse compatible Bazel build and test results
- See executed work, reused results, and resource usage
- $3 introductory compute credit, valid for 7 days
LLM optimization is included within the stated monthly allowances. No model credits are included. Hosted CI requires repository qualification and owner-approved payment setup. Existing subscription customers can manage their current agreement in workspace billing.
The one-time $3 compute credit expires after 7 days. Usage beyond the credit, or after expiry, is billed monthly at the displayed rate. Card required to activate compute. Cache/storage limits and any separately quoted network or storage charges are confirmed before activation. Taxes may apply.
Move from GitHub Actions
Keep GitHub.
Move the Actions.
Keep your repositories, pull requests, and reviews on GitHub. Import supported workflow YAML into Compress, share compatible Bazel results, and see what actually ran.
- 01Connect and inspect.
Choose a repository. See supported workflows, missing secrets, and compatibility gaps.
- 02Import and verify.
Start with manual execution. Validate a real run on the current revision before switching.
- 03Switch when it passes.
Enable Compress triggers and disable the matching GitHub workflow. Keep a rollback path.
Keep your workflow YAML. Switch after verification.
Start with one supported Linux workflow. See compatibility gaps before moving execution. Qualified workspaces can activate metered compute with a card and a one-time $3 credit valid for 7 days. Provider LLM usage is separate.
Your CI bill. With the math shown.
Compare monthly compute for a 2-vCPU Linux x64 workload. No monthly Compress fee.
GitHub Actions$162.00/mo
$0.006/min, less $18.00 available credit.
Blacksmith$108.00/mo
$0.004/min, less 3,000 available free minutes.
Compress$90.00/mo
$0.003/min equivalent. $0 platform fee.
$72.00 lessper month vs GitHub · 44% lower compute bill
$18.00 lessper month vs Blacksmith · 17% lower compute bill
Adjust reuse and runtime assumptions
Start with equal runtime and no assumed reuse. Use measured timings when you have them. A test-cache hit percentage is not a percentage reduction in your whole job.
1× means equal billable runtime; 0.5× means half the runtime. Estimated Compress allocation: 30,000 two-vCPU equivalent minutes.
Ongoing monthly compute estimate. Use GitHub billed job minutes; GitHub rounds each job up to a minute. Excludes the one-time introductory credit, taxes, provider LLM charges, and separately agreed storage/network overages. Hardware, RAM, concurrency, and performance differ. Compress meters allocated vCPU-seconds from setup through verified cleanup. Queue time and results reused without a worker allocation incur no compute charge.
Check your workflow compatibilityBefore you switch
Start with one workflow.
Do I need to move off GitHub?
Keep your repositories, PRs, and reviews. The guided import checks supported Linux workflow YAML, required secrets, and execution compatibility. Run the current revision manually before switching triggers. Your existing GitHub workflow stays available until you choose to disable it.
Does every repository need Bazel?
No. Supported workflows can run without Bazel. Sharing Bazel build and test results between an agent and CI requires Bazel targets, matching declared inputs and a qualified runtime, plus authorized cache access. Changed inputs and checks without reusable evidence still execute.
Will Compress always cost less?
Your bill depends on runtime, allocated resources, reusable work, and the allowances you already have. Free standard public GitHub runners or Blacksmith’s free minutes can cost less for smaller workloads. Use the calculator, then compare a real workflow before moving more.
What is included in the seven-day offer?
A first eligible subscription receives a one-time $3 compute credit, valid for 7 days from activation. A card and usage authorization are required. Compute beyond that credit or after expiry is paid. This is not unlimited free execution.
Can I use LLM optimization without buying CI?
Yes. Connect your own provider keys and use the included optimization allowance. Providers bill your model usage directly. Hosted CI is optional and requires repository qualification; activating it does not raise the LLM allowance.
Start with either. Bring both together.
Make every
cycle count.
Less repeated compute. More room to build.