> ## Documentation Index
>
> Fetch the complete documentation index at: https://lithi.ai/llms.txt
>
> Use this file to discover all available pages before exploring further.

# Keep your agents. Make your usage go further.

Delegate suitable code, data, document and media work to Lithi at token pricing. Depending on the work, this can preserve your subscription allowance or reduce comparable API costs.

No account or installation needed to review.

[Get a free estimate](/compute#compute-assessment)

[Apply for Compute access](/compute#compute-assessment-actions)

## Crowdsourced AI compute, powered by people.

Mac owners choose when their computers take part in the Lithi Agent Network.

### Connect your agent.

Use Lithi skills with a supported MCP or HTTP client. The Compute Kit guides preparation, approved delegation and result checks.

### Use nearby Macs when possible.

Lithi matches work with approved Macs that meet its requirements, including capacity in your area when available.

### Support real people.

50% of eligible settled task revenue goes to the provider pool for Macs that help complete accepted work. The rest runs and grows the network.

[Learn about Lithi Connect](/connect)

## Your agent plans. Lithi coordinates the work.

Lithi splits suitable work across approved Macs, checks it against your agent’s requirements and combines the results.

### What you bring

#### Text & Data

Input: Records to organize

Task: Structure the records

Result: Structured records

Records to organize. You choose the task—here, structure the records. Lithi plans suitable batches, matches available Macs, checks and combines their work, and returns structured records. Nearby matching depends on your requirements and available capacity.

#### Documents & Images

Input: Documents to sort

Task: Sort documents by category

Result: Sorted documents

Documents to sort. You choose the task—here, sort documents by category. Lithi plans suitable batches, matches available Macs, checks and combines their work, and returns sorted documents. Nearby matching depends on your requirements and available capacity.

#### Audio & Video

Input: Recordings to transcribe

Task: Transcribe with timestamps

Result: Timestamped transcript

Recordings to transcribe. You choose the task—here, transcribe with timestamps. Lithi plans suitable batches, matches available Macs, checks and combines their work, and returns timestamped transcript. Nearby matching depends on your requirements and available capacity.

#### Code & Tests

Input: A code change with its callers, types and test evidence

Task: Review the change against a rubric

Result: Findings with source locations and repair suggestions

A code change with its callers, types and test evidence. You choose the task—here, review the change against a rubric. Lithi plans suitable batches, matches available Macs, checks and combines their work, and returns findings with source locations and repair suggestions. Nearby matching depends on your requirements and available capacity.

#### Models & Evaluation

Input: Answers and a rubric

Task: Compare answers against the rubric

Result: Scored comparisons

Answers and a rubric. You choose the task—here, compare answers against the rubric. Lithi plans suitable batches, matches available Macs, checks and combines their work, and returns scored comparisons. Nearby matching depends on your requirements and available capacity.

[Got Questions? Ask Lithi](/compute)

## Know when to use Lithi—and when to wait.

### Preserve allowance

Delegate suitable tasks to keep more allowance for work your agent still handles.

### Compare API costs

Compare equivalent output and quality, including preparation, review and work retained by your agent.

### Consider timing

Compare your deadline with your allowance reset and available capacity. Waiting may be better.

[Check usage fit](/docs/compute/usage-fit)

[Check cost fit](/docs/compute/cost-fit)

## Prepare locally. Approve what leaves.

Your agent uses the Compute Kit to prepare and check work on your machine.

### Select what matters

Your agent identifies the fields each task needs. The Kit leaves unnecessary data out of the batch.

### Prepare the data

Remove or redact sensitive fields and replace permitted identities with private codes. This is pseudonymization. The matching list stays on your machine.

### Check the complete batch

Scan for secrets and sensitive content, then re-check the prepared request. Flag missing checks and unclear identity matches for review.

### Review and approve

Resolve flagged issues and approve the batch before transfer. Blocked work and work awaiting review stay local.

[How the Kit prepares your data](/docs/compute/security/preparation)

## Pay for input and output tokens.

Per million tokens · USD

| Input | Output |
| --- | --- |
| $0.05 | $0.10 |

The rate includes normal platform retries and hedges, default verification, and delivery or repair.

[See pricing and examples](/pricing)

## Prepaid credit. A clear record of your work.

### Your credit records

Your paid purchase invoice records the credit you buy. Your monthly usage statement shows accepted work and credit used, with zero prepaid amount due. There is no automatic reload or overage debt.

### Your impact report

This separate document shows your accepted work, the Macs that helped, and the electricity used and its emissions, each marked measured, estimated or unavailable.

### Your local contribution

See the share of covered work completed in your region and confirmed payments to participating providers there.

[How to read your report](/docs/compute/impact-report)

[How we measure impact](/trust/impact)

## Find out where Lithi fits your work.

Give your agent the review prompt. It uses the guides and work context you authorize to identify up to five suitable opportunities.

### Review the fit

Check supported tasks, data requirements and how your agent connects.

### Compare your options

Compare allowance use, timing and total cost.

### Choose your next step

Keep the recommendation, refine a sample task or request Compute access.

[Get a free estimate](/compute#compute-assessment)

[Apply for Compute access](/compute#compute-assessment-actions)

No account. No installation. No paid work during the review.

## Common questions

### Will Lithi increase my usage limits? {#usage-limits}

No. Lithi is built to help you hit limits less often by moving suitable work that would otherwise consume the same allowance or API usage. Your provider still controls its limits and reset times. Planning and reviewing delegated work can still use your agent’s allowance.

### When does my agent recommend Lithi? {#agent-recommendation}

Your agent checks usage pressure, reset timing, task fit, data requirements and total cost using supported readings or information you provide. A usage threshold starts a review; paid work requires approval. Continuing where you are or waiting may be better.

### Will using Lithi always cost less? {#cost-comparison}

No. Compare equivalent work and quality, including preparation, review and work your agent still performs. Preserving an included allowance is not automatically a cash saving. When the evidence is incomplete, your agent should explain what is missing rather than invent a saving.

### What can participating Macs read? {#mac-data-access}

A participating Mac can read its assigned, prepared input. The Kit limits what is shared and can replace permitted identities with private codes; the matching list stays with you. These steps do not guarantee anonymity or confidential computing.

### Do I need to install the Kit to review Lithi? {#review-without-installing}

No. Your agent can review the public guides without an account or installation. Using the supplied Kit requires an approved account and compatible interface. Reading the guides is not a source-code audit and does not install, upload or run work.

### How are results checked? {#result-checks}

Lithi checks outputs against agreed schemas, rules, comparisons or review steps. Your agent also checks coverage, exceptions and the returned evidence before using a result. Verification shows what was checked; it does not make every answer correct.

### How do I try a small task? {#small-task}

Ask your agent to assess a small task and its permitted data. After account approval and setup, confirm the account and Work tools, prepare a sample, review the quote and permissions, then run once with the required funding and approval. Check the status, result, coverage and evidence before scaling.

[Use Compute with your AI](/docs/compute/clients/frontier-agent-setup)

[Compute Docs](/docs/compute)