Roo Code Guide: What It Is and Who It’s For

Roo Code shows up in a lot of 2025 “best open-source coding agent” roundups. Some of those posts are still useful. Many are quietly wrong about what you should install in 2026.

This guide is deliberately conservative. Roo Code was a real, widely used VS Code coding agent that forked from Cline, leaned into modes and autonomy, and then — per public GitHub status — archived its primary repository on 15 May 2026. Community forks and a separate hosted direction exist in the aftermath. We are not going to invent a tidy price list or pretend the marketplace page on your machine is still upstream-maintained.

If you need a living OSS agent we actually review end-to-end, start with Cline and the Cline AI guide. If you need a commercial agent ceiling, read Claude Code and Claude Code pricing. Landscape: complete AI coding guide and coding hub.

Verified against public repository/archive signals and secondary reporting as of 2026-08-05. Open-source status can change again — re-check GitHub and official domains before you standardise.

The short answer

1. Roo Code ≈ Cline lineage, different product philosophy (more modes / autonomy emphasis in its active period).
2. Primary repo archived 15 May 2026 — treat “current default OSS agent” claims with suspicion unless a living fork is named.
3. Default recommendation in 2026: Cline for OSS BYOK; commercial tools when you want support and a seat ceiling.
4. Community forks (e.g. ZooCode-style efforts mentioned in public posts) are evaluate carefully, not automatic replacements.
5. No invented pricing for archived IDE builds; BYOK costs still apply if you run old bits against live APIs.
6. Hero/logo note: we have no Roo logo in-catalogue; this article’s hero uses Cline as the honest related mark.

What people meant when they said “Roo Code”

During its active extension life, Roo Code was described consistently across independent writeups as:

• a VS Code extension,
• open source (Apache-2.0 licensing appeared on the public project),
• BYOK (bring your own model keys — same economic family as Cline),
• a fork of Cline that diverged toward custom modes, multi-step agent workflows, and often more autonomous defaults.

That last point is the product idea in one sentence: Cline-like agent plumbing with a more “team of modes” workflow on top.

It was never magic intelligence separate from the model you attached. Like every BYOK agent, quality tracked whatever frontier or local model you paid for.

Relationship to Cline (be precise)

Fork does not mean:

• settings files always compatible,
• security patches flow automatically,
• marketplace updates continue forever,
• “install either” is responsible advice for a company.

If your team standardised on Roo in 2025, your 2026 job is migration planning, not nostalgia.

The May 2026 archive (why old rankings break)

Public GitHub state for RooCodeInc/Roo-Code shows the repository archived by the owner on 15 May 2026, i.e. read-only. Archive messaging in the ecosystem points people at:

• Cline — the project Roo originally forked from,
• community forks started to preserve Roo-style behaviour (ZooCode is the name that appears repeatedly in third-party posts; verify the org and commit activity yourself).

When a coding agent archives, three clocks start:

1. Model adapter drift — providers change APIs; unmaintained clients lag.
2. Prompt injection / tool-risk research — fixes do not land.
3. Marketplace trust — copies and mirrors proliferate; supply chain gets messier.

Running an archived agent against production repositories is a risk decision, not a lifestyle brand.

The pivot narrative (hosted vs IDE)

Around the archive, public commentary describes the Roo organisation pivoting toward a hosted / Slack-first / cloud agent direction (Roomote / roomote.dev style branding appears in secondary sources; redirects from older Roo domains are reported unevenly).

What we will not do here:

• invent a current monthly price for that hosted product,
• claim feature parity with Devin Cloud or Claude Code,
• assert uptime, SOC2, or training policies we have not audited on-page.

If you are buying hosted autonomous coding, compare living vendors with primary sources: Cognition’s Devin story, Claude Code, Cursor, etc. Do not assume the old Roo extension and a new hosted SKU are the same tool with a coat of paint.

Who Roo was for (when it was alive)

Honest historical fit:

• Developers who liked Cline’s BYOK but wanted more mode machinery.
• People experimenting with multi-agent style roles inside VS Code.
• Power users willing to trade some restraint for configurability.

Poor fit even then:

• Anyone who wanted zero API console work.
• Teams needing vendor support SLAs on the extension.
• Pure autocomplete buyers (Copilot / Cursor Tab).

That persona map still helps you read old blog posts without installing anything.

Who should care in 2026

You should read this page if

• A tutorial told you to “just install Roo Code” and you want current status.
• You are auditing why two agents share DNA (Cline vs Roo forks).
• You maintain a fork or internal patch set and need language for leadership.
• You are cleaning a tool sprawl spreadsheet full of 2025 logos.

You should probably install something else if

• You want a maintained OSS agent → Cline, Cline AI guide, Cline vs Claude Code.
• You want predictable paid polish → Claude Code or Cursor.
• You want free Tab → Devin Desktop free tier notes in best free AI coding tools.
• You want assistant-first help → best AI coding assistants.

Risks specific to archived / forked agents

Supply chain

Download only from repositories and marketplace publishers you can verify. Archived projects attract impersonators. Prefer hashes and org history over a random “Roo” VSIX on a forum.

Security surface

Agents that run shell and edit files are already high risk when maintained. Unmaintained tool-calling stacks are worse. Pair any experiment with:

• disposable keys + hard spend caps,
• non-production repos first,
• no cloud credentials in the environment,
• mandatory human approval on network and shell.

Policy and compliance

“It’s open source” is not a data-processing agreement. BYOK still sends code context to whatever model host you configured. Procurement cares about that host, not the editor plugin’s licence alone.

Knowledge drift

Stack Overflow answers and YouTube installs age badly in this category. If the thumbnail says 2025 and the title says “best agent,” check the repo commit date before you spend an afternoon.

Community forks: how to evaluate without romance

When a popular tool archives, forks appear. Some are good-faith stewardship. Some are drive-by renames.

Evaluation checklist:

1. Commit recency measured in days, not stars alone.
2. Named maintainers with history, not only an anonymous org.
3. Issue response on security-tinged tickets.
4. Clear lineage (what was forked, what still merges from Cline).
5. Licence still OSI-approved and actually matching the code.
6. Default autonomy — more modes can mean more ways to do damage.
7. Migration docs from Roo configs, if any.

If a fork fails this checklist, use Cline or a commercial agent. Your job is shipping software, not cosplaying package maintainer for free.

Pricing: what we can and cannot say

For live BYOK economics, use our Cline review numbers as the closest documented analogue: extension free for individuals; working developers often discuss $20–70/month model spend with no ceiling unless capped.

Migration paths (practical)

From Roo-era setup → Cline

1. Export or note your model provider settings (keys stay in the provider).
2. Install current Cline from official channels.
3. Re-apply spend caps immediately.
4. Re-teach project conventions (root markdown / rules files).
5. Run one identical ticket; compare diff quality and cost.
6. Uninstall abandoned extensions once parity is good enough.

From Roo-era setup → Claude Code

1. Accept paid floor (Pro ~$20; no free Code) — pricing guide.
2. Write CLAUDE.md.
3. Follow how to use Claude Code.
4. Expect less model shopping, more product integration.

From Roo-era setup → Cursor

1. Import VS Code muscle memory.
2. Use Tab for assistant work; Agent for tickets.
3. Read how to use Cursor AI.

Staying on a fork

Only with explicit ownership: who patches, who reviews releases, who holds the keys. “Everyone uses it” is not ownership.

How Roo fits the wider agent market

The complete AI agents guide argues that coding is where agents work best because compilers and tests give feedback. Roo participated in that wave. The market lesson from its archive is sharper:

Open-source agent UX is easy to fork and hard to staff forever.

Commercial tools fail by pricing opacity. OSS tools fail by maintenance cliffs. You need a strategy for both.

Comparison snapshot (decision aid)

What PromptHive covers (and does not)

We maintain a full tool review for Cline, not a separate Roo SKU page. That is intentional: our catalogue prefers products readers can still responsibly adopt. This article exists so searchers landing on “Roo Code” get status honesty and a path to living tools — not a zombie feature matrix.

We also will not pretend multi-vendor logos we do not ship; the hero uses Cline because that is the truthful neighbour in /public/logos.

Related reading

• Cline review · Cline AI guide
• Cline vs Claude Code
• Claude Code review · How to use Claude Code
• Best free AI coding tools
• Best AI coding assistants
• Complete AI coding guide · Best AI coding tools
• Complete AI agents guide
• Devin AI review — commercial cloud agent contrast

What “modes” meant in practice (historical)

During Roo’s active period, writeups repeatedly emphasised mode-driven workflows: different postures for planning, coding, debugging, or other roles, sometimes closer to a multi-agent theatre inside one extension. That is attractive if you like explicit phase changes. It is dangerous if “mode” becomes a synonym for “fewer approvals.”

If you recreate Roo-like behaviour on Cline or a fork today:

1. Keep plan artifacts human-readable before tool execution.
2. Do not map “architect mode” to unattended shell.
3. Log which mode produced which diff for audit.
4. Prefer fewer modes with clear norms over a dozen half-documented personas.

Modes are UX. Permissions and git are safety.

How to read a 2025 “Roo vs Cline” article in 2026

Apply this filter:

If the article does not mention the May 2026 archive, treat its install steps as historical.

Internal RFC outline (if your company still has Roo)

Copy and adapt:

1. Current state: who has Roo installed; which model keys; which repos.
2. Risk: archived upstream; no security backports expected.
3. Options: Cline; Claude Code; Cursor; vetted fork with named owner.
4. Pilot: two engineers, two weeks, same tickets, cost + revert metrics.
5. Decision date: uninstall deadline for non-approved extensions.
6. Exception process: only with security sign-off and spend caps.

Without a deadline, zombie extensions live forever.

Why we still wrote this page

Search demand does not die when a repo archives. People will keep typing “Roo Code” into Google for years. Leaving them with only outdated listicles is how they install malware-adjacent VSIX files or burn money on unmaintained tool chains.

PromptHive’s job here is status honesty + redirection, not nostalgia ranking.

Bottom line

Roo Code mattered. It proved developers wanted Cline’s open agent idea with different workflow defaults. It also demonstrated the failure mode of the category: when the maintainers pivot, the extension becomes archaeology.

In 2026, do not build your engineering system on an archived default. Use Cline if you want the living OSS BYOK line. Use Claude Code or Cursor if you want a product team behind the agent. Treat forks as explicit projects, not background utilities.

If a bookmark still says “install Roo,” update the bookmark before you update the repo.