GitHub Copilot coding agent and Ivy Tendril both turn a task description into a pull request, but they do it in different places and with different amounts of control. Copilot coding agent is assigned a GitHub issue, works in a GitHub-hosted environment built on GitHub Actions, and opens a pull request for you to review. Ivy Tendril is a local-first desktop application that runs the coding agent of your choice on your own machine, through a fixed sequence of stages with two human checkpoints, one at the plan and one at the diff, with many tasks executing in parallel in separate git worktrees. Pick Copilot coding agent if you want the least setup and you are committed to GitHub and the models it offers. Pick Ivy Tendril if code must stay on your machine, if you want to review a plan before execution, choose the agent and model per task, and run many tasks at once.
What each approach is for
GitHub Copilot coding agent is a hosted feature of GitHub Copilot. You assign an issue to Copilot, and the agent clones the repository into a GitHub-hosted environment, makes changes, runs what it can, and opens a pull request. You review that pull request the way you would review one from a colleague. There is nothing to install and nothing to run locally. The agent is Copilot, and the models are the ones GitHub makes available, which change over time. At the time of writing, access is tied to a Copilot subscription; check GitHub's pricing page for current terms.
Ivy Tendril is a desktop application for macOS, Windows, and Linux from Ivy Interactive AB in Stockholm. It also runs headless with tendril --web. It orchestrates AI coding agents from a plan through execution, verification, and review to a pull request. The application is free and source-available under the Functional Source License (FSL-1.1-ALv2). Pro and Enterprise plans add team features, on-prem hosting, SSO, verification imports from CI, and support. Tendril is agent-agnostic: it runs Claude Code, OpenAI Codex CLI, GitHub Copilot CLI, Google Gemini CLI, OpenCode, or any other CLI agent, and you bring your own API keys.
The two are not mutually exclusive. Tendril can run GitHub Copilot CLI as one of its agents, and it ingests GitHub issues through webhooks. See the GitHub integration docs.
Comparison
| Aspect | Ivy Tendril | GitHub Copilot coding agent |
|---|---|---|
| Where code runs | Your machine, or your own server in headless mode | GitHub-hosted environment (GitHub Actions) |
| Human checkpoints | Two: plan review and diff review | One: pull request review |
| Parallelism | Many plans at once, each agent in its own git worktree | One task per assigned issue; you can assign several issues |
| Agent choice | Claude Code, Codex CLI, Copilot CLI, Gemini CLI, OpenCode, any CLI agent | Copilot |
| Model choice | Any provider you hold keys for | Models GitHub offers |
| Cost | Free application; you pay your model provider directly; tokens tracked per plan and per job | Part of Copilot subscription terms at the time of writing |
| Where plans, memory, and logs live | On your machine | GitHub |
| Agent memory | Promptware memory per lifecycle stage, persists across runs | Repository instruction files, at the time of writing |
| Repository host | GitHub for issues and pull requests | GitHub |
| License | Free, source-available (FSL-1.1-ALv2) | Proprietary |
Where the code runs
With Copilot coding agent, the repository is cloned into an environment GitHub operates. That is the main convenience: nothing to install and nothing to keep running. It is also the main constraint. If your tests need a local database, a licensed toolchain, or a private network, you configure that inside the Actions environment, and the work happens on infrastructure you do not operate.
With Ivy Tendril, agents run on your machine, each in a git worktree. The only external calls are to the LLM API you selected and to GitHub. Plans, agent memory, and execution logs stay on disk. Tendril does not proxy your traffic, and there is no training on your code. Teams that want the same setup on a shared server run tendril --web, and Cloudflare Quick Tunnels let you steer a run from a phone. Read more in local-first AI development.
Human checkpoints
Copilot coding agent has one checkpoint: the pull request. You write an issue, the agent interprets it, and the first time you see how it interpreted it is when the pull request appears. If the agent misunderstood the scope, the cost of that misunderstanding is a full run plus your review time.
Ivy Tendril has two checkpoints. The lifecycle is: idea or ticket, create plan, draft, human reviews the plan, then Expand, Split or Update it, execute in worktrees, verify (tests, lint, diff), human reviews the diff, pull request, merge. The plan review is where you catch a wrong interpretation before any code is written. You can comment inline on a draft and the plan is rewritten. Read more in from GitHub issue to pull request.
Parallelism and the main branch
Both tools keep agent work off your main branch until a pull request is reviewed. In Tendril, many plans execute at the same time, each agent in its own git worktree on its own branch. The Dashboard shows the status of every plan, the Jobs view streams agent output, and the Review view shows the diff and the verification results in two tabs. Ivy reports that its own team went from roughly 10 to more than 100 pull requests per day after adopting the workflow. See the article on git worktrees for parallel AI agents.
With Copilot coding agent, you can assign several issues, and each one gets its own branch and pull request.
Agent, model, and cost control
Copilot coding agent uses Copilot and the models GitHub offers. The cost is part of your subscription, which is predictable per seat but leaves the model choice, and the price per task, to GitHub.
Ivy Tendril lets you switch the agent or model per plan without changing the workflow. Bring your own keys for Anthropic, OpenAI, Google, OpenRouter, Vercel AI Gateway, Cloudflare, Scaleway, NVIDIA, and others. Token usage and cost are tracked per plan and per job, and the Dashboard shows cost KPIs and a trend chart. Routine plans can run on a cheaper model and difficult ones on a more capable model, and the difference shows in the numbers.
When GitHub Copilot coding agent is the right choice
- Your team already has Copilot seats and wants to use them with no additional tooling.
- Your task volume is low and issues are well scoped, so a single checkpoint at the pull request is enough.
- You do not want to manage API keys or model providers.
- You accept the repository being cloned into a GitHub-hosted environment, and your builds already run in Actions.
When Ivy Tendril is the right choice
- Code must stay on your machines, for regulatory, contractual, or policy reasons.
- You want to review the plan before an agent executes it, and to split or update plans before execution.
- You want to run ten or twenty tasks in parallel and review a queue of diffs.
- You want to choose the agent and model per task, and to change them as providers change.
- You want agent instructions and memory that persist and improve across runs, through promptware.
- You want a free, source-available tool, with paid team features when you need them. See pricing.
Frequently asked questions
Can I use GitHub Copilot inside Ivy Tendril?
Yes. GitHub Copilot CLI is one of the supported agents. You can run it for some plans and Claude Code or Codex CLI for others, and the workflow, checkpoints, and cost tracking stay the same.
Does Ivy Tendril require GitHub?
Pull requests and issue ingest go through GitHub, so a GitHub repository is the expected setup. Planning, execution, verification, and diff review all happen on your machine, before anything is pushed. The walkthrough from GitHub issue to pull request shows the full path.
What does Ivy Tendril cost?
The application is free under FSL-1.1-ALv2. Pro is $59 per user per month and adds team features, on-prem hosting, SSO, verification imports from CI, and support. Enterprise plans are available. You pay your model provider directly for tokens; Tendril does not proxy that traffic.