Last updated: August 2026 · Written after two of this segment's leading tools shut down within a month of each other
Open-source AI coding tools promise something proprietary IDEs cannot: code you can audit, a project you can fork if the vendor disappears, and no lock-in to a single company's roadmap. That promise got tested directly in mid-2026, when two of the segment's most prominent tools — Continue.dev and Roo Code — shut down within four weeks of each other, for two completely different reasons.
This guide covers the open-source AI coding landscape as it actually stands today: which tools are genuinely active, what happened to the ones that aren't, and where their users went.
Continue.dev was acquired by Cursor in an acqui-hire announced June 16, 2026 — Cursor's third acquisition in 18 months — and discontinued shortly after. This is the risk open-source advocates usually associate with proprietary tools, not open-source ones: a well-funded competitor buys the team and the product goes away. The Apache 2.0 code remains technically forkable, but no one has stepped up to maintain it as of this writing. Full details: Continue.dev: What Happened.
Roo Code shut down May 15, 2026, for an entirely different reason: its own team decided IDE-based coding assistance was the wrong bet and pivoted to Roomote, a cloud-agent product. This is a startup-strategy risk, not an acquisition risk — open source didn't prevent it, but it did make the outcome better than Continue.dev's: Roo Code's code lives on through Kilo Code, an actively maintained fork with an official migration path. Full details: Roo Code: Where to Go.
The pattern worth noticing: both teams, independently, pointed their users toward the same tool when they shut down — Cline. That is not a coincidence of marketing; it reflects where the actual center of gravity in this segment has settled.
Cline is currently the most-adopted open-source AI coding agent, with over 5 million installs and roughly 58,000 GitHub stars. Released under Apache 2.0, it runs as a VS Code extension — your editor, keybindings, and existing plugins stay exactly as they are.
Cline is agent-first, not autocomplete-first: you describe a task, it plans and executes across files and terminal, with approval required at each step by default. Its MCP marketplace is the most developed of any tool in this category — one-click server installation, and the ability to generate a new MCP server from a natural-language description when nothing existing fits. Checkpoints (session-level rollback via a shadow git repository) add a safety net independent of your normal git workflow.
Pricing is BYOK: the extension is free, you pay API costs directly to your chosen model provider (typically $5–20/month at moderate use with Claude Sonnet 4.6, or $0 with a local Ollama model).
Full configuration reference: Cline Rules guide. Direct comparison against the proprietary leader: Cursor vs Cline.
| Cline | |
|---|---|
| License | Apache 2.0 |
| Installs / stars | 5M+ installs, ~58K GitHub stars |
| Interface | VS Code extension (agent only, no autocomplete) |
| Pricing | Free + API costs, $0 with Ollama |
| Best for | Most open-source use cases — the default starting point |
Aider predates Cursor, Claude Code, and most of this category — Paul Gauthier started it in 2023, and it has grown to roughly 40,000+ GitHub stars and several million PyPI installs under Apache 2.0. It runs entirely in the terminal against a local git repository, with no IDE dependency at all.
Its defining feature is the repository map: a compressed, tree-sitter-generated structural representation of your codebase that lets Aider identify relevant files on large projects without loading everything into context. Every successful edit is auto-committed with a generated message, producing a clean, auditable git history by default.
A note on release pace: Aider's development has genuinely slowed in 2026 — the most recent data available shows no new PyPI release since February 2026, several months of gap as of this writing. This is better read as a mature, stable tool than an abandoned one (the community remains large and active, and the core functionality is not broken), but it is worth knowing before you standardize tooling around it, and worth periodically checking aider.chat for the current release cadence yourself.
Full configuration reference: Aider Rules guide. Direct comparison against Cline: Cline vs Aider.
| Aider | |
|---|---|
| License | Apache 2.0 |
| Stars / installs | 40K+ GitHub stars, millions of PyPI installs |
| Interface | Terminal / CLI only |
| Pricing | Free + API costs |
| Best for | Git-native workflows, CI/CD automation, large codebase refactors |
Kilo Code is a fork of Roo Code that rebuilt on a new architecture (OpenCode server) and reached general availability April 2, 2026 — a few weeks before Roo Code itself shut down. It automatically reads a former Roo Code project's .roomodes and .roo/rules/ configuration on first startup, making it the closest thing to a drop-in continuation for anyone specifically attached to Roo's multi-agent, custom-mode workflow.
Kilo runs across VS Code, JetBrains, and as a standalone CLI on a shared session model — broader IDE coverage than Roo Code itself had, and broader than Cline's VS Code-only free tier.
Full migration details: Roo Code: Where to Go.
| Kilo Code | |
|---|---|
| License | Open source |
| Interface | VS Code, JetBrains, CLI |
| Pricing | Free + API costs |
| Best for | Former Roo Code users; anyone wanting JetBrains support in a free agent |
Zed is a different category from the three tools above — a complete, standalone code editor built in Rust from scratch, open source, rather than an extension added to VS Code. It is the open-source answer to "I want a full IDE like Cursor, not just an agent bolted onto my existing editor."
AI is BYOK across 15+ providers with no markup, and the Agent Client Protocol (ACP) lets you connect external agents — including Claude Code — natively inside the editor rather than through a bolted-on integration. Performance is a first-class design goal, not an afterthought: 0.12s startup, GPU-accelerated rendering, and roughly a third of Cursor's idle memory footprint.
Full comparison against Cursor: Cursor vs Zed.
| Zed | |
|---|---|
| License | Open source (GPL/Apache mix) |
| Interface | Standalone native editor |
| Pricing | Free (editor) / $10/mo Pro; AI is BYOK either way |
| Best for | Developers wanting a full open-source IDE, not just an extension |
Was one of the most configurable open-source options, with genuine JetBrains support and per-role model assignment (a fast local model for autocomplete, a frontier model for chat). Acquired by Cursor and shut down; final release June 19, 2026; GitHub repository now read-only. The Apache 2.0 code can technically be forked, but no community fork has emerged as a maintained successor as of this writing. If you are still running a pinned installation, see the archived configuration reference — it is preserved for exactly that use case.
Was a popular Cline fork known for custom multi-agent "modes" and orchestration. Shut down by its own team's choice, not acquired — they concluded cloud agents (their new product, Roomote) were the better bet than IDE extensions. Unlike Continue.dev, this one has a genuine living successor: Kilo Code, covered above, plus Cline as the official recommended path for users who don't need mode-based orchestration specifically.
It's worth being honest about what open-source licensing did and did not protect against here, since it's easy to overstate the guarantee.
What it didn't do: prevent either product from being discontinued. Apache 2.0 code being technically forkable is not the same as someone actually maintaining a fork — that requires a team willing to do the ongoing work, which is a separate and harder thing to guarantee than the license itself.
What it did do, in Roo Code's case specifically: make a genuine continuation possible. Kilo Code exists, has an official migration guide, and reached GA before Roo's shutdown — because Roo's code being open meant a separate team could build directly on it rather than starting over. Continue.dev's code is equally open, but as of this writing no comparable fork has emerged, illustrating that the license is necessary but not sufficient — someone still has to show up and do the work.
The practical lesson for choosing a tool in this category: community size and momentum matter as much as the license itself. Cline's scale (5M+ installs) makes it a safer long-term bet not because it's open source, but because a project with that much adoption has a much larger pool of people who could step in if the core team ever stepped back — the same safety property Kilo Code demonstrated for Roo Code's userbase, just pre-emptively rather than reactively.
| Tool | License | Interface | Free tier cost | Best for |
|---|---|---|---|---|
| Cline | Apache 2.0 | VS Code extension | $0 + API ($0 w/ Ollama) | Default starting point, MCP power users |
| Aider | Apache 2.0 | Terminal / CLI | $0 + API | Git-native, CI/CD, large codebases |
| Kilo Code | Open source | VS Code, JetBrains, CLI | $0 + API | Former Roo Code users, JetBrains |
| Zed | Open source | Standalone editor | $0 (BYOK for AI) | Full IDE replacement, performance |
| Apache 2.0 | — | Discontinued June 2026 | ||
| Open source | — | Discontinued May 2026 — see Kilo Code |
You want the safest, most-adopted open-source option with no specific other requirement → Cline. It's the default for a reason — scale, MCP maturity, active development.
You want a full IDE, not an extension, and prefer native performance → Zed. The only option here that replaces your editor entirely rather than adding to it.
You live in the terminal and want git-native, auditable AI edits → Aider. Just be aware of its slower 2026 release cadence if you're standardizing team tooling around it.
You were on Roo Code specifically and want the closest continuation → Kilo Code for feature parity and JetBrains support, or Cline if you're open to leaving the multi-mode workflow behind.
You were on Continue.dev specifically → Cline is the closest equivalent for VS Code privacy/BYOK work; there is no direct JetBrains-plus-free equivalent left in this category — Cline's JetBrains support requires the Enterprise tier.
You want maximum privacy with zero external transmission → Cline or Aider, both with a local Ollama model. See AI Coding Tools: Privacy & Security Compared for the full architectural breakdown.
Cline, for most use cases — it has the largest install base, the most developed MCP ecosystem, and was the tool both Continue.dev and Roo Code's own teams pointed their users toward when they shut down. For a full open-source IDE rather than an extension, Zed is the closest equivalent to what Cursor offers structurally.
Not for new adoption. It was acquired by Cursor and discontinued in June 2026. An existing installation may continue to function since it hasn't been remotely disabled, but there will be no further updates or security patches. See Continue.dev: What Happened for the full story and an archived configuration reference.
No — it shut down May 15, 2026. Its own team recommended Cline as the migration path; Kilo Code, a fork of Roo, offers closer feature continuity with automatic configuration migration. See Roo Code: Where to Go for the full migration guide.
Yes, though its release pace has slowed in 2026 relative to earlier in the project's life — the most recent available data shows several months since the last published release as of this writing. The community and install base remain large, and the tool remains functional and widely used; check aider.chat directly for the current release status before standardizing team tooling around it.
Coincidence of timing more than a shared cause — the two shutdowns had entirely different drivers. Continue.dev was an acquisition (Cursor buying the team). Roo Code was a strategic pivot by its own founders (betting on cloud agents over IDE extensions). Both landing within about a month of each other in 2026 reflects how fast-moving and consolidating this specific product category has become, not a common underlying failure.
Partially. The license guarantees the code remains legally forkable, but it does not guarantee anyone will actually maintain a fork — that takes a team choosing to do the work, which happened for Roo Code (via Kilo Code) but has not happened for Continue.dev as of this writing. A large, active user and contributor base is a better practical predictor of continuity than the license alone.
Likely yes, in the short term — neither extension has been remotely disabled, so an existing installation should keep functioning. But neither will receive further updates, security patches, or compatibility fixes as AI provider APIs change, so both should be treated as a migration-in-progress rather than a stable long-term choice. See the dedicated guides for Continue.dev and Roo Code migration steps.