Skip to content
Browse documentation

Projects & context

Context, usage, results, and history

Interpret what Vopit stores, sends, and reports.

Vopit composes working context from the request, attachments, project evidence, handoffs, ticket state, and durable project facts. It can use repository maps and JS/TS symbol information, with text fallback when Graphify is unavailable. Native CLIs manage their own internal context.

Tune context carefully

  • Focused working memory and Graphify are enabled by default. The context ceiling defaults to 64,000 characters and can range from 16,000 to 400,000.
  • Request narrow reads or explicit expansion when an agent needs more source. Repeat reads can be reused; archives preserve older material.
  • Durable project memory is evidence, not truth. Treat stale sources as stale and ask the agent to verify current files.

Read usage reports accurately

Usage views can show input, output, cached input, cache-write, and reasoning tokens plus calls, duration, and estimated cost. Reasoning tokens are included in output. A missing metric is unknown, not zero; configured per-million prices make an estimate rather than a provider invoice. Character counts and pruning percentages are not billed-token savings.

Inspect and export records

Results include activity, timeline, modified files, usage, and ticket information. Request and event JSON exports and structured logs can contain project data. Each project stores its team, settings, connections, and layout in `.vopit/config.json`, and its history and revisions in `.vopit/history.sqlite`. Connection secrets remain in the application credential store and are associated with the workspace.

Cache is part of input, not an extra input total

Reported input includes cached reads and cache writes; do not add these to input again. Reasoning is included in output. For configured API rates, Vopit prices ordinary input, cached input, cache writes and output separately. An unknown cache price falls back to the configured input rate and is marked incomplete. This is an estimate, not an invoice. Account/CLI usage has no Vopit dollar estimate; its reported tokens do not convert directly into subscription quota.

One native invocation represents an entire native CLI stage and can sum many internal model calls. A large reported input total is not one giant Vopit prompt. Invocation details show the latest 60 entries; aggregates retain all saved usage. Thinking groups public activity and tool evidence; it does not expose the model's private reasoning.

All documentation