Inside the workspace

Follow the work, from question to change.

Use an AI agent with the project in view: its files, the task plan, changed-file diffs and terminal output. Keep the evidence close to the decisions you make.

Context

Read and search the project

Open a local repository, browse its tree, read relevant files and search for the code behind a behaviour. Build the conversation around the project itself.

Agent

Plan and follow the task

Give the agent a goal and constraints. Follow its plan and progress, ask a follow-up or stop the run when the direction needs to change.

Changes

Inspect file edits and diffs

Prepare focused file changes with the agent, then compare the changed code. Keep the scope small enough to understand what changed and why.

Terminal

Check with real commands

Run commands, builds and tests from the project workspace. Read the resulting terminal output instead of relying on an unsupported success claim.

Continuity

Revisit conversations and runs

Return to saved conversations and run history to recover the task context. The account portal provides project records and history controls.

Account

Organize access without leaving it implicit

Use personal or company workspaces, invitations and member roles. Manage API keys and device access in the portal. Company billing and paid seats are planned.

Beyond chat

Focused subtasks

Break a larger task into smaller pieces. Delegate a focused subtask and follow its own conversation and progress.

Sources and connected tools

Use available web search, open a source or connect a supported MCP tool. Connections and network access depend on your setup and permissions.

UGP-Solo on your PC

Run a compatible model locally on Windows. Private local mode disables Solo network tools; enable selected network tools separately when needed. Model installation and suitable hardware are required.

Three useful starting points

Small, well-defined tasks are easier to understand and verify. Start with a question about the code, then expand the scope when you have the context.

Understand unfamiliar code

Ask where a request enters the application, which files handle it and what assumptions the code makes. Follow the referenced files before choosing a change.

Fix a reproducible problem

Describe what happens and what you expect. Ask for the smallest relevant edit, inspect the diff and run a check that covers the original problem.

Improve test coverage

Choose one behaviour or edge case. Let the agent inspect existing test conventions, prepare a focused test and run the relevant project command.

Choose the model and review the result

  • Supported local models can run inference on your computer. Cloud models and network tools may send relevant context to their providers.
  • Local-model requests have no UGP request-count cap. Hardware requirements and model capability still apply.
  • External-provider expense limits are separate from the subscription price and are not included credits.
  • AI output requires review. Check permissions, changed files and command output before committing or deploying.