Lithi Compute Kit

The Kit gives your existing agent a harness for supported work. In familiar terms, a harness is the surrounding path that helps your agent prepare permitted input, connect an approved interface, request a quote, submit approved work and handle its result. Your agent keeps its own model, context, planning, edits, builds, tests and final decisions. Skills teach that path; they do not train model weights or grant permission.

Participating Macs on the Lithi Agent Network process admitted tasks with local models. Your agent checks what Lithi returns and decides what to keep. A quote prices a declared scope; transfer permission and execution approval remain separate.

From task to checked result

Your agent assesses the task and keeps its context
             ↓
The Kit prepares permitted input locally
             ↓
An approved account and interface carry the request
             ↓
Review the quote, transfer permission and execution approval
             ↓
Lithi Agent Network processes the work; status and result return
             ↓
Your agent checks the result and continues its work
             ↓
An authorized monthly report explains recorded work and participation

Your agent supplies the outcome, permitted information and acceptance checks. The Kit selects a workflow, prepares the smallest useful input and connects the supported client surface. A compatible agent needs a supported MCP or API transport and approved account access; a chat interface does not connect automatically. An executing Mac reads the plaintext of its granted piece; encryption does not turn a provider Mac into a confidential-computing enclave.

For a code review, ask your agent to prepare a permitted diff, its relevant callers and recorded test evidence. Lithi can return findings tied to source; your agent checks each finding, makes any accepted repository edits and runs the build and tests in your environment. Start with the Code & Tests guide to prepare a small review.

Choose the next guide

Start with the work menu when you need to define a task. Read Kit contents to choose a supplied artifact path, the agent guide for a selective reading bundle, or Kit review to compare data boundaries, timing and cost before access.

The five public information families describe input shape: Code & Tests, Text & Data, Documents & Images, Audio & Video, and Models & Evaluation. The eight workflow guides are Lithi Text & Data (llm-agent-work), Lithi Audio & Video (creator-media), Lithi Code & Tests (code-review-testing), Lithi Data & Evaluation (data-evaluation), Lithi Documents & Search (documents-retrieval), Lithi Translation (translation-localization), Lithi Apple ML (apple-ml) and Lithi Image Analysis (vision). These labels organize reading; the selected service still decides accepted format, operation, limits and output. Vision has no domain skill.

Connect and review

Claude Code, Codex and Cursor can use portable skills and MCP where the selected host supports those surfaces. Host compatibility does not prove access to local files, an alias key or a result destination. Use the host guide for current details.

Informational pricing is USD 0.05 per million input tokens and USD 0.10 per million output tokens. Use those figures to compare a measured workload; a signed quote sets the amount for a specific scope. For subscription usage, compare the measured allowance the task would consume with the planning and review that remain. For API billing, compare like-for-like current charges with Lithi cost, retained agent work and included or excluded overhead.

Public pages are the learning path. The optional assessment JSON provides structured page and pricing references; it does not grant access to a release, keys, an archive or a work tool. Read Kit review before Compute access.

Read the monthly report

With reporting:read permission, your agent can interpret an authorized Monthly Compute impact report through the reporting API. It can also read an approved report file. Reporting is a separate scope and interface; it is not a sixth Work tool. The report can distinguish accepted_work_macs, physical_activity_macs and provider_accounts. Treat those as different populations rather than adding them together, and use the report's availability labels to understand what its evidence establishes. Allocation remains separate from confirmed payment, and measured or estimated energy remains separate from a claimed saving.