Skip to content
ReviewRamble

The weekly brief

AI, reviewed, in your inbox.

  • New reviews of tools, models, MCP servers and repos, with the verdict up front
  • Re-checked prices and ratings when something changes
  • New analysis pieces as they are published

Models, agents, MCP servers and the tools actually worth your time.

One email a week. Unsubscribe with the link in any brief. How we use your email

The best open-source AI coding agents in 2026

The best open-source AI coding agents right now are OpenCode, OpenAI Codex CLI and Cline, out of 38 we have reviewed. Each is ranked by our rating, with its price, our verdict and who it suits.

By ReviewRamble editors · Updated Nov 18, 2026 · How we review

How we ranked: by our rating out of 5, from reviews built on each product's own pricing, documentation and policies plus independent benchmarks where they exist, each checked on the date shown in the review. We take no affiliate commission.

  1. 1

    A terminal agent you can point at any model and audit line by line, provided you keep its local server patched.

    Best for: Developers who want a vendor-neutral terminal coding agent.

    Price: Free and open source (MIT); you pay your model provider. Optional OpenCode Go costs $10 or $40 a month; Zen is pay-as-you-go from $20 top-ups; Enterprise is per seat, price on request.

    Read our OpenCode review
    4.6/5
  2. 2

    For anyone already paying for ChatGPT, Codex CLI is a well-sandboxed terminal agent at no extra cost, on allowances OpenAI only estimates.

    Best for: ChatGPT subscribers who want a terminal coding agent without another bill.

    Price: CLI is open source (Apache 2.0). Usage comes with ChatGPT Plus ($20/month), Pro ($100 to $500/month), Business ($20/user/month annual) and Enterprise, or pay-per-token with an API key.

    Read our OpenAI Codex CLI review
    4.5/5
  3. 3

    Cline is now the default open-source agent for VS Code, more because its rivals stopped than because it changed, and its 2026 security record says to treat it as software with real attack surface.

    Best for: VS Code users who want an open-source agent and control over which model they pay for.

    Price: Free and open source (Apache 2.0); you pay model providers or Cline at cost. ClinePass for open-weight models $9.99/month. Enterprise custom.

    Read our Cline review
    4.4/5
  4. 4

    herdr is tmux rebuilt around the question you actually ask when five agents are running, which one is waiting on me, and it answers it well.

    Best for: Terminal-first developers running several coding agents across projects or machines who want them to outlive the client and to see at a glance which one needs an answer.

    Price: Free and open source (Apache-2.0). You bring your own agent subscriptions or API keys; no paid tier is published.

    Read our herdr review
    4.4/5
  5. 5

    OpenHands has turned from a single autonomous coder into an open-source switchboard for many, which makes it more useful to teams and easier to misconfigure.

    Best for: Engineering teams that want to run and automate several coding agents on their own infrastructure.

    Price: Agent Canvas free to self-host (MIT), with your own model key. Teams $499/month for up to 10 users (extra seats $50/month, 5,000 sessions a month); Enterprise custom.

    Read our OpenHands review
    4.4/5
  6. 6

    Plain git worktrees with the friction removed, and for anyone running agents in parallel that friction was the whole problem.

    Best for: Developers running several coding agents in parallel on one repo who want worktrees handled without adopting a whole agent dashboard.

    Price: Free and open source (MIT or Apache-2.0).

    Read our Worktrunk review
    4.4/5
  7. 7

    Pi is the coding agent for people who would rather shape their own tool than learn someone else's, and it has the engineering discipline to be trusted with that job.

    Best for: Developers who want a lean terminal agent they can extend, script or embed in their own product.

    Price: Free and open source (MIT). You pay your model provider or log in with a supported subscription through OAuth.

    Read our Pi review
    4.3/5
  8. 8

    A free, agent-neutral control room for parallel coding agents, held back only by how fast it changes underneath you.

    Best for: Developers who run several coding agents at once and want them isolated, tracked and reviewed in one place.

    Price: Free and open source under MIT. You bring your own agent subscriptions or API keys.

    Read our Orca review
    4.2/5
  9. 9

    T3 Code is the remote control for the coding agents you already pay for, and the one that lets you walk away from your desk while they work.

    Best for: Developers who run several coding agents and want one open surface, including a phone, to drive and watch them.

    Price: Free and open source (MIT). You bring your own agent subscriptions (Claude Code, Codex, Cursor and others).

    Read our T3 Code review
    4.2/5
  10. 10

    Crush is a carefully built terminal agent with sensible permission controls, but its licence is not open source and Charm's own model plans changed shape two weeks ago.

    Best for: Developers who want a polished, model-agnostic terminal agent with language-server context and fine-grained tool permissions, and who will not resell it.

    Price: Free to use under FSL-1.1-MIT (source-available; each version becomes MIT two years after release). Bring your own API keys or local models, or pay Charm's Hyper: individual plans $10 to $200/month with weekly credits at 5 cents each, Business plans $30 to $220 per user with zero data retention, or prepaid credit from $5.

    Read our Crush review
    4.1/5
  11. 11

    A vendor-neutral agent for automating work across many systems, with coding as one job among several and autonomy switched on from the first prompt.

    Best for: Developers and power users who want an open, extensible agent for coding and general automation.

    Price: Free and open source (Apache 2.0). You pay your model provider or run local models.

    Read our goose review
    4.1/5
  12. 12

    Open Code Review fixes the specific ways coding agents fail at review, skipped files and drifting line numbers, and it trades some recall for comments worth reading.

    Best for: Engineering teams that want precise, line-anchored AI review in CI or the terminal on a model and endpoint they control.

    Price: Free and open source (Apache-2.0). You pay for the LLM endpoint you configure; delegation mode uses your coding agent's existing model instead.

    Read our Open Code Review review
    4.1/5
  13. 13

    Archon turns "please remember to run the tests" into a loop the agent cannot leave until they pass, which is the right idea, shipped by a project still breaking its own interfaces every few weeks.

    Best for: Engineering teams that have a delivery process they trust and want coding agents to follow it every time, with a human approving only the decisions that need one.

    Price: Free under MIT. No paid edition; you pay for the Claude Code or Codex subscriptions or API usage the workflows consume.

    Read our Archon review
    4.0/5
  14. 14

    Kilo Code is now the main heir to Roo Code's mode-based workflow, with real JetBrains support, but it ships faster than it stabilises.

    Best for: Developers who want a mode-based open-source agent in VS Code or JetBrains.

    Price: Extension and CLI free (MIT). BYOK or pay-as-you-go gateway credits (5% processing fee); optional Kilo Pass $19, $49 or $199/month with bonus credits; Teams $15/user/month; Enterprise custom.

    Read our Kilo Code review
    4.0/5
  15. 15

    A benchmark harness so plain that its results measure the model, which is exactly why it feels bare as a daily coding tool.

    Best for: Researchers and engineers benchmarking models or learning how coding agents work.

    Price: Free and open source (MIT). You pay your model provider.

    Read our mini-swe-agent review
    4.0/5
  16. 16

    Claude Squad is the lightest way to run parallel coding agents if you already live in tmux, and it deliberately does little beyond that.

    Best for: Terminal-first developers who want to run a few coding agents in parallel without adopting a desktop app or a new workflow.

    Price: Free and open source under AGPL-3.0. You bring your own agent subscriptions or API keys.

    Read our Claude Squad review
    3.9/5
  17. 17

    Daintree is the most thought-through cockpit for running several coding agents at once, with refreshingly honest docs, but it is a small project with 87 stars, led by one developer, whose code is written by AI agents and which accepts no outside pull requests.

    Best for: Developers who already pay for one or more coding-agent CLIs and want to run several tasks in parallel with a proper review step before anything merges.

    Price: Free and open source (Apache 2.0), no account. You pay for the agent CLIs' own subscriptions or API usage; Daintree does not relay or bill model traffic.

    Read our Daintree review
    3.9/5
  18. 18

    OpenQodex builds the second opinion a coding agent should get before every push, and its evidence checks are sound, but its own real-pull-request numbers show how much it still misses.

    Best for: Developers on macOS or Linux who let Claude Code or Codex write most of their changes and want an independent, evidence-checked review before each push without a new account or bill.

    Price: Free and open source (Apache-2.0), with no account or key of its own. Reviews run on your existing Claude Code or Codex login; the GitHub Action's full review needs an Anthropic API key.

    Read our OpenQodex review
    3.9/5
  19. 19

    xtool puts iOS builds within reach of a Linux agent, but Apple's sign-in still needs a human and a plugged-in phone.

    Best for: Developers and agent setups on Linux or Windows that need to build and install SwiftPM iOS apps on a real device without a Mac.

    Price: Free and open source (MIT). Password login works with any Apple ID; API-key login and App Store or TestFlight distribution need the paid Apple Developer Program ($99 a year).

    Read our xtool review
    3.9/5
  20. 20

    Gas City turns Gas Town's agent factory into readable configuration on top of Beads, but out of the box it still runs Claude at maximum effort with no permission prompts.

    Best for: Teams that want several coding agents to plan, build and review work in parallel and are willing to set permissions and models per agent.

    Price: Free and open source (MIT). Agents bill to your own Claude, Codex or Gemini logins or API keys. Gas City, Inc. offers a commercial Beads Team Server to design partners; prices not published.

    Read our Gas City review
    3.8/5
  21. 21

    A sensible bet that agents need a backend they can inspect and change themselves, held back for now by its youth and by leaning on Vercel and OpenRouter for two of its parts.

    Best for: Solo developers and small teams who let a coding agent build full-stack prototypes and want the agent to manage the backend too.

    Price: Apache 2.0 and free to self-host. InsForge Cloud: Free; Pro $25/month or $20/month billed yearly, plus compute from $5/month per project with a $10 monthly credit; Enterprise custom.

    Read our InsForge review
    3.8/5
  22. 22

    Ovis is the phone remote herdr does not ship itself, careful about who gets in, but young, fast-moving and run by one person.

    Best for: herdr users who want to read, steer and approve their agents from a phone or another computer and are willing to run Tailscale or their own HTTPS route.

    Price: Free and open source (MIT), with no account. Needs herdr, your own always-on computer and your own agent subscriptions.

    Read our Ovis review
    3.8/5
  23. 23

    Qwen Code has outgrown the Gemini CLI it was forked from, but without its free tier it is mainly a good front end for Alibaba's $50 Coding Plan, and its daemon ships with no password.

    Best for: Developers who want a feature-rich, Apache-licensed terminal agent and are happy to pay Alibaba or another provider for the model.

    Price: Free and open source (Apache 2.0). Models cost extra: Alibaba Cloud Coding Plan Pro $50/month (up to 90,000 requests a month, interactive use only), pay-as-you-go Token Plans, any third-party API key, or local models. The free Qwen OAuth tier ended on 15 April 2026.

    Read our Qwen Code review
    3.8/5
  24. 24

    A thoughtful agent workbench from a JetBrains incubator team that gets worktree isolation right, though it is still at version 0.3 and tied to the pi engine.

    Best for: Developers who run several coding-agent tasks in parallel and want worktree isolation and a PR workflow in one open-source app.

    Price: Free and open source (Apache-2.0); model use is billed by JetBrains AI or your own provider.

    Read our ThinkRail review
    3.8/5
  25. 25

    Cheap DeepSeek tokens and a harness you can rebuild piece by piece, wrapped in unaudited preview code that sends your session logs to DeepSeek by default.

    Best for: Developers on DeepSeek's API who want an agent they can rebuild from plugins, run on a machine or VM with nothing sensitive on it.

    Price: Free and open source (MIT); you pay for model usage. On DeepSeek's API at peak, deepseek-flash costs $0.30 per million uncached input tokens and $1.20 per million output tokens, deepseek-v4-pro $1.32 and $3.96; off-peak rates are half.

    Read our DeepSeek Harness review
    3.7/5
  26. 26

    LoopForge gives coding agents a real delivery process, with separate reviewers and human sign-off, but its defaults suit a trusted team on a private network and nothing wider.

    Best for: Engineering teams of trusted colleagues who want coding agents to follow the same plan, review, test and sign-off steps on every change, on their own servers.

    Price: Free and open source under the MIT licence; model usage is billed by each provider on your own keys.

    Read our LoopForge review
    3.7/5
  27. 27

    oh-my-hermes gives Hermes the planning, parallel lanes and honest status reports of a serious coding harness, but it is five months old, on its fourth major version and still has no published benchmark showing it helps.

    Best for: Hermes Agent users who want planned, parallel, multi-model coding runs with clear evidence of what actually passed.

    Price: Free under the MIT licence; you pay the model providers Hermes calls. No paid edition is published.

    Read our oh-my-hermes review
    3.7/5
  28. 28

    A careful design for letting an agent near production credentials, built by a team that went to Vercel in July and has barely touched the code since.

    Best for: Small platform teams that want an infrastructure agent which never sees production secrets in plain text.

    Price: Agent free under Apache 2.0 with your own API keys. Hosted plans: Hacker $15/month (3 apps, $10 credits), Builder $45/month (10 apps, $35 credits), Team $450/month (50 apps, $350 shared credits), Enterprise custom.

    Read our Stakpak review
    3.7/5
  29. 29

    Wasp gives coding agents an unusually clear map of a full-stack app, at the cost of betting on a pre-1.0 framework that still breaks things between releases.

    Best for: Solo founders and small teams building React and Node.js SaaS apps with a coding agent who accept pre-1.0 upgrade churn.

    Price: Free and open source (MIT); no paid tier. You pay for hosting and any model provider you use.

    Read our Wasp review
    3.7/5
  30. 30

    Aider's commit-per-edit discipline still holds up, but a tool with one point release in a year is a habit to keep rather than one to start.

    Best for: Existing users who want a minimal, git-centred pair-programmer in the terminal.

    Price: Free and open source (Apache 2.0). You pay your model provider or run local models.

    Read our Aider review
    3.6/5
  31. 31

    An ambitious harness that takes open models seriously, with headline benchmarks that are its own and telemetry you have to switch off.

    Best for: Developers who want to run coding agents on cheap open models across several repositories and are happy with early-preview software.

    Price: Free and open source (Apache-2.0). You pay your model provider; default per-task spend cap is $5.

    Read our CodeAF review
    3.6/5
  32. 32

    A thoughtful process for making coding agents prove their work at each step, wrapped in more surface area than most small teams will want to learn.

    Best for: Teams that already run coding agents and want a visible, rule-based path from request to reviewed commit.

    Price: Free and open source under the MIT licence; model and agent costs come from the providers you connect.

    Read our Routa review
    3.5/5
  33. 33

    Vibe Kanban still works well as a local board for agent tasks, but with its company gone and no release since April, it is a risky thing to build a team habit on.

    Best for: Solo developers who like planning agent work on a board and are comfortable running, or forking, a community-maintained tool.

    Price: Free and open source under Apache-2.0. Paid cloud subscriptions ended when bloop shut down in April 2026; you bring your own agent subscriptions or API keys.

    Read our Vibe Kanban review
    3.4/5
  34. 34

    Gas Town shows a Mayor and a crew of agents can work a backlog unattended, but it burns tokens fast, skips every permission prompt and has handed its future to Gas City.

    Best for: Experienced Claude Code users with a large token budget who want to watch a fully autonomous multi-agent workflow and can run it in a container.

    Price: Free and open source (MIT). Every agent bills to your own Claude, Codex, Copilot or other subscription or API key, and a fleet multiplies that bill.

    Read our Gas Town review
    3.3/5
  35. 35

    A capable open-source agent that Google has already moved on from, and hard to justify adopting in late 2026.

    Best for: Existing Gemini Code Assist customers and Vertex AI teams who want an open-source terminal agent until their contracts run out.

    Price: Open source (Apache 2.0). The free Google-account tier ended on 18 June 2026; use now needs a paid Gemini API key, Vertex AI or an existing Code Assist licence.

    Read our Gemini CLI review
    3.3/5
  36. 36

    SwarmForge is a well-argued experiment in running coding agents like a disciplined team, but its missing licence and permission-bypassing defaults keep it a thing to study rather than adopt.

    Best for: Engineers studying how to structure multi-agent coding pipelines, who can run it in an isolated machine and do not need to redistribute it.

    Price: Free to download, but the repository has no licence file, so reuse rights are not granted. You pay for the agent backends (Claude, Codex, Copilot or Grok) it runs.

    Read our SwarmForge review
    3.3/5
  37. 37

    A neat idea for drawing agent pipelines on top of OpenCode, but a two-month-old, one-person fork that edits real files with auto-approved permissions is a hobby tool for now.

    Best for: OpenCode users who want to experiment with drawn, reusable multi-agent coding pipelines on a sandbox repository.

    Price: Free and open source (MIT); models are free on OpenCode's shared tier or billed by your provider.

    Read our OpenFlow review
    3.2/5
  38. 38

    Claw Code is a piece of GitHub history with a working CLI attached, and its own maintainers point you elsewhere for real work.

    Best for: Engineers who want to read how a Claude Code-style harness fits together, not to adopt one.

    Price: Free and open source (MIT). You pay the model provider whose API key you configure.

    Read our Claw Code review
    2.5/5

Sources

The main official source behind each ranked review; every review lists its full set.

More guides