ProPR: the self-hosted alternative to Copilot cloud agent

Copilot cloud agent takes an issue, works in a GitHub-managed sandbox, and opens a pull request on GitHub's meter. ProPR runs the same agents on your server and subscriptions, with plan approval, review, and Ultrafix.

GitHub Copilot

Who meters the work

Copilot cloud agent bills every task in GitHub AI credits and keeps session logs on GitHub. ProPR bills nothing: your provider charges your plan, and every task record lives on your infrastructure.

Copilot cloud agent pathAssign the issue → GitHub cloud sandbox → pull request → @copilotEvery task draws GitHub AI credits, and the session logs live on GitHub.

ProPR pathPlan → isolated workspace → GitHub PR → follow-up commitNo credits. Your provider bills your plan, and the task record stays on your server.

Switch from Copilot cloud agent to ProPR.

What you keep: your GitHub repositories and review process, and the Claude or ChatGPT plan, which ProPR runs through Claude Code or Codex.

The same ideas, ProPR's names

Copilot cloud agent	ProPR
Assign the issue to Copilot	Add the `AI` label to the issue
Plan when the prompt asks for one	Plan, approved on every task
GitHub cloud sandbox	Isolated workspace on your server
`@copilot` or "Fix with Copilot"	Any plain PR comment; `/fix` for review findings
Copilot models, plus the hosted Claude and Codex agents	Claude Code, Codex, Antigravity CLI, Mistral Vibe, OpenCode
GitHub AI credits per token	Your provider plan, no markup
Session logs on GitHub	Task record and usage log on your server

From zero to the first pull request

Install ProPR, or skip the server

Run npm install -g propr-cli and propr setup on a Linux Docker host (quickstart). ProPR Hosted runs a dedicated instance for you when you would rather not run one.

Add the same repository and sign in your agent

Open the web UI, sign in with GitHub, and add the repository. Sign in Claude Code or Codex with the plan you already hold.

Start the first task and approve the plan

Add the AI label to the issue you would have assigned to Copilot, then approve the plan ProPR posts.

Review the pull request in GitHub

Review the diff the way you review any pull request. A plain comment on it becomes the next commit on the same branch; /review posts scored findings and /fix applies them. The Ultrafix playbook loops review and fix until the score you set.

What stays with Copilot cloud agent: Copilot cloud agent needs no server and its compute comes from GitHub; ProPR runs on a Linux Docker host you own or on a ProPR Hosted plan.

Copilot cloud agent lives where issues, branches, and reviews already live. Assign an issue to Copilot, or to the Claude and Codex agents GitHub hosts in public preview, and it works in a GitHub-managed cloud sandbox and opens a pull request. Mention @copilot on that PR, or batch review comments with "Fix with Copilot", and it pushes commits to the branch. Ask for a plan in the prompt and it drafts one for approval first. Every task draws GitHub AI credits priced per model and token, on top of a paid Copilot plan from $10 to $100 per month for individuals; Actions-based runs can use self-hosted runners, but orchestration, logs, and billing stay on GitHub.

ProPR moves the whole loop onto your server. Each task starts with a plan you approve, runs Claude Code, Codex, Antigravity CLI, Mistral Vibe, or OpenCode in isolation, and ends in a pull request. A second model reviews it, Ultrafix keeps fixing until the checks pass, and plain PR comments become follow-up commits on the same branch. The plan you already pay for covers the tokens; ProPR adds no meter, and the usage log shows every call.

The LLM log: each model call with its tokens and cost, plus the subscription capacity it used: Claude session and weekly windows, the Codex weekly window, the Antigravity Gemini weekly limit, and none for OpenCode — ProPR's LLM log: every call with its tokens and cost, drawn from the Claude, Codex, or Antigravity subscription window you already pay for.

Copilot cloud agent fits best when...

The organization already pays for Copilot and wants agent runs inside GitHub's own product.
GitHub-managed compute and AI-credit billing are acceptable, and nothing needs to leave GitHub.
Copilot's models plus the hosted Claude and Codex agents cover the work.

ProPR fits best when...

Agents run on your server, on subscriptions you already hold, with no per-task credits.
A plan precedes every implementation and a second model reviews the result.
Plans, logs, task history, credentials, and usage live in your own deployment.

GitHub-hosted agent runs versus self-hosted PR orchestration.

Question	Copilot cloud agent	ProPR
Open source (Apache 2.0)	No	Yes
Where the agent runs	GitHub cloud sandbox by default; Actions runners optional	Your server
Agents you can assign	Copilot models, plus Claude and Codex in public preview	Claude Code, Codex, Antigravity CLI, Mistral Vibe, OpenCode
How agent work is billed	Paid Copilot plan plus GitHub AI credits per model token	Your provider plans or API keys; no markup
Plan approval before code	When the prompt asks for a plan	Every task
Review by a second model	No	`/review` and Ultrafix
PR comments become follow-up commits	Yes: `@copilot` or "Fix with Copilot"	Any plain comment, any agent
Task logs and credentials self-hosted	No	Yes

Verified November 2026: About Copilot in the cloud, third-party coding agents, using the cloud agent on GitHub, plan mode changelog, Copilot plans, AI-credit billing.