#1★ TOP PICK
Cline
The open-source agent that lives in the editor you already have.
92
OPEN SOURCEApache-2.0SELF-HOSTLOCAL-FIRST
Cline (Apache-2.0, by Cline Bot Inc.) is an autonomous coding agent that installs as a VS Code extension, a JetBrains plugin or a CLI — so you keep your editor, your extensions and your keybindings and simply add the agent. It is model-agnostic: bring your own key for Anthropic, OpenAI, Google, Bedrock and 100+ others, or point it at Ollama or LM Studio and run the whole loop on your own machine. Cline sells optional prepaid credits and an enterprise tier, but nothing about the extension requires an account, and with your own key there is no Cline backend in the request path. At roughly 4.81 million VS Code installs and 65.2k GitHub stars it is the most widely adopted open-source coding agent, and because it is published on Open VSX you can also install it inside Cursor itself.
⌁ Cline itself runs on anything that runs VS Code — the hardware question is only about the model. Cline's own docs put local models at 16–32 GB RAM for small quantised models, 32–64 GB for mid-size coding models and 64 GB+ for larger ones with expanded context. As a concrete anchor, Ollama's default qwen3-coder tag (30B total, ~3.3B active) is a 19 GB download with a 256K context window, so roughly 24 GB of VRAM — or a 32 GB+ unified-memory machine — is a sensible floor for comfortable agentic work. Below that, a cloud API key is usually the better call, and a local model still won't match a frontier cloud model on hard multi-file tasks.
Strengths
- +Keeps your existing IDE, extensions and keybindings — the smallest possible change from Cursor
- +Bring-your-own-key with no account required: 100+ providers, or fully local via Ollama / LM Studio
- +Plan/Act split and diffs you approve before anything is written, plus MCP tool support
- +The most-adopted open agent in the category: ~4.81M VS Code installs, 65.2k stars, commits daily
Trade-offs
- −No inline Tab-style autocomplete — Cline is an agent, not a completion engine, and this is its most-requested missing feature
- −You manage your own API billing; long agent runs on a frontier model can cost more than a $20 Cursor seat
- −Cline's own prepaid-credits path is the promoted onboarding — bring-your-own-key takes a couple of extra settings
- −Running a capable model locally is a real hardware ask (see the hardware note)
Free and open source (Apache-2.0) — you pay only your model provider. Optional prepaid Cline credits and a paid enterprise tier exist; bring-your-own-key needs no account. #2
opencode
The terminal-native agent with the largest open-source community in the category.
93
OPEN SOURCEMITSELF-HOSTLOCAL-FIRST
opencode (MIT) is a coding agent that runs in your terminal, with a desktop app in beta and IDE extensions alongside it. It is deliberately provider-agnostic — 75+ providers via the AI SDK and Models.dev, including Ollama, LM Studio and any OpenAI-compatible local endpoint — and it ships two built-in agents: a full-access `build` agent and a read-only `plan` agent for exploring a codebase before anything is allowed to change it. It is now the most-starred project in this space at roughly 191k GitHub stars, maintained by Anomaly after the repository moved from `sst/opencode` to `anomalyco/opencode`. The optional opencode Zen gateway sells curated, benchmarked models pay-as-you-go and includes some free ones; using your own provider keys stays free.
⌁ The client is a single binary and runs comfortably on modest hardware, including over SSH. Running the model locally is the demanding part: budget ~24 GB VRAM or a 32 GB+ unified-memory machine for a 30B-class quantised coding model, and expect a capability gap against a frontier cloud model on multi-step agent work.
Strengths
- +MIT and provider-agnostic by design — 75+ providers, plus Ollama, LM Studio or any OpenAI-compatible local endpoint
- +The read-only `plan` agent lets you scope work before the agent can touch a file
- +Terminal-first, so it works over SSH on a server or a modest machine
- +The largest community in the category and daily commits
Trade-offs
- −Terminal-first is a genuine adjustment coming from an editor-integrated agent; the desktop app is still beta
- −No inline autocomplete — a different tool for a different job than Cursor's Tab
- −Ownership moved from SST to Anomaly; the MIT licence means no practical risk, but it's worth knowing
- −Large configuration surface (providers, models, agents) — more to learn than picking from a dropdown
Free and open source (MIT) — you pay your own provider. The optional opencode Zen gateway is pay-as-you-go, with some free models; your own API keys cost nothing extra. #3
OpenHands
Self-hosted, sandboxed coding agents — the whole stack runs on infrastructure you control.
91
OPEN SOURCEMITSELF-HOSTLOCAL-FIRST
OpenHands (MIT, formerly OpenDevin) is the option for people whose actual requirement is "this has to run on infrastructure I control." It is a self-hosted control centre for coding agents: a browser-based Agent Canvas plus a backend server you start locally or on your own VM, with agents working inside a sandboxed environment that has its own shell, filesystem and browser. Agents can edit code, run commands, call APIs and browse the web — real engineering tasks rather than completions. The project sells a managed OpenHands Cloud and licenses a self-hosted enterprise edition, but the core is MIT and self-hosting costs nothing beyond your own model usage. At 82.5k stars it is the most established self-hosted agent platform available.
⌁ Plan for a machine that can comfortably run Docker plus a sandbox container alongside your normal workload — 16 GB RAM is a practical minimum and 32 GB is more comfortable, whether that's a workstation or a small VM. If you also want the model in-house, that is a separate ~24 GB VRAM budget on top.
Strengths
- +Genuinely self-hosted: server, sandbox and workspace all sit on your own infrastructure
- +Sandboxed execution — the agent gets a container, not unrestricted access to your laptop
- +Browser UI suits longer autonomous runs and reviewing what an agent actually did
- +MIT-licensed, 82.5k stars, committed to daily
Trade-offs
- −The heaviest setup here — Docker and real RAM, not a two-minute install
- −Long autonomous runs consume a lot of tokens; set a budget before you let one loose
- −Multi-user, RBAC and the richer integrations live in the paid Cloud and enterprise tiers
- −Not an in-editor experience — you leave your IDE to drive it, and there's no inline completion
Free and open source (MIT) to self-host — you pay only your model provider. Managed OpenHands Cloud and a licensed enterprise self-host are the paid tiers. #4
Aider
Git-native pair programming in the terminal — every change lands as a commit.
89
OPEN SOURCEApache-2.0SELF-HOSTLOCAL-FIRST
Aider (Apache-2.0) is the minimalist of this group: a Python CLI that edits files in your repository and commits each change with a sensible message, so your review tool is `git diff` and your undo is `git revert`. It connects to almost any LLM including local ones — `aider --model ollama_chat/<model>` with `OLLAMA_API_BASE` set — and it has no account, no backend and no vendor anywhere in the path. The honest caveat is cadence: the most recent PyPI release (0.86.2) landed on 12 February 2026 and the latest commit on GitHub is 22 May 2026. It still works, and 47.8k people star it, but in a field where models change monthly that is a noticeably slower pulse than the others here — which is why it scores well on sovereignty and lower on maintenance.
⌁ The lightest option here: a Python CLI that will happily run on a low-powered machine or a small VM while calling a cloud model. If you want the model local too, the same ~24 GB VRAM guidance applies for a 30B-class quantised coding model, and Aider's Ollama docs are worth reading first because of the default context-window behaviour.
Strengths
- +Git-native: every edit is a commit you can diff, revert or cherry-pick — the best audit trail of the group
- +No account, no backend, no vendor in the request path at all
- +The lightest client here — a Python CLI that runs anywhere, including over SSH on modest hardware
- +Clear, well-documented local-model path via Ollama
Trade-offs
- −Development has slowed sharply — last PyPI release 12 February 2026, last commit 22 May 2026
- −Terminal only: no editor UI and no inline completion
- −Ollama defaults to a 2k context window and silently discards overflow, so local setups need deliberate configuration (Aider's docs cover it, but it is a footgun)
- −A leaner feature set than the newer agents — no sandboxing or built-in browser tooling
Free and open source (Apache-2.0) — you pay only your model provider. No account, no hosted tier, no credits. #5
GitHub Copilot
The polished commercial option at half the price — and for many readers the honest answer.
28
SOURCE-AVAILABLEProprietary (hosted service; MIT-licensed VS Code chat client)
If what actually prompted the comparison is price rather than openness, GitHub Copilot is the straightforward answer: Pro is $10/month against Cursor Pro's $20, and there is a real free tier with 2,000 code completions a month for individual developers who don't have Copilot through an organisation. It does inline completion, chat, agent mode and a CLI, and it installs into the IDE you already use rather than asking you to adopt a different editor. Worth knowing precisely: the Copilot Chat client for VS Code is MIT-licensed and its code now lives in the main VS Code repository, but the service and models behind it are proprietary and hosted. There is no self-host path and no local-model option — that's the trade, stated plainly, and it is the right trade for plenty of people.
Strengths
- +Half the price of a Cursor Pro seat, with a genuinely free tier to test on
- +Nothing to run or configure — it installs into the IDE you already use
- +Keeps inline Tab-style completion, which none of the open agents above provide
- +Very low switching cost — your code is just files, so uninstalling leaves nothing to migrate
Trade-offs
- −Fully hosted and proprietary: no self-hosting, no local models, no offline use
- −Chat, agent and CLI usage draws on monthly credits; only completions are unlimited on paid plans
- −You are choosing a Microsoft/GitHub-operated service, with whatever that means for your data policy
- −The Sovereignty Score of 28 reflects a closed hosted service — a transparency lens, not a verdict on quality
Free tier (2,000 completions/month); Pro $10/mo; Pro+ $39/mo; Max $100/mo; Business $19/seat/mo; Enterprise $39/seat/mo. Verified on GitHub's own plans documentation 2026-07-29.