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 tools for coding agents in 2026

The best open-source tools for coding agents right now are ast-grep, ccusage and agent-device, out of 41 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

    ast-grep is the structural search tool to hand a coding agent before it reaches for sed, as long as nobody mistakes its syntax matching for semantic analysis.

    Best for: Developers and coding agents that need precise, repeatable searches and codemods across a large codebase without writing a compiler plugin.

    Price: Free and open source (MIT).

    Read our ast-grep review
    4.5/5
  2. 2

    ccusage is the meter every heavy agent user should run, provided they read its dollar column as what the work would have cost on the API, not what they paid.

    Best for: Anyone who wants to track or compare token use across agent sessions.

    Price: Free and open source (MIT).

    Read our ccusage review
    4.4/5
  3. 3

    If your agents write mobile code, agent-device is how they check it runs, and it already has the device juggling and CI replay that one-off mobile MCP servers lack.

    Best for: Mobile and React Native teams that want coding agents to verify their own changes on simulators, devices or a device cloud and attach the evidence to pull requests.

    Price: Free under the MIT licence; device clouds are billed by their providers, and Callstack sells team QA services on request.

    Read our agent-device review
    4.3/5
  4. 4

    Memory for coding agents that starts with everything you have already done, because it reads the transcripts instead of waiting for the agent to save notes.

    Best for: Developers who switch between several coding agents and want each one to know what the others already fixed, without a hosted memory service.

    Price: Free under the MIT licence; no hosted or paid edition.

    Read our deja-vu review
    4.2/5
  5. 5

    Plannotator fixes the weakest moment in agent coding, reading a long plan in a terminal, and does it locally with unusually candid privacy notes.

    Best for: Developers who use plan mode in Claude Code, Codex or similar agents every day and want to review plans and diffs with the care they give a pull request.

    Price: Free and open source under MIT or Apache-2.0. Workspaces, a hosted team layer, is on a waitlist with no published price. AI questions inside reviews use your own agent or model provider.

    Read our Plannotator review
    4.2/5
  6. 6

    A careful, well-documented security scanner, provided you treat it as a model reading your source under your own account rather than an isolated sandbox.

    Best for: Security and platform teams that want model-driven code review in CI and can give each report a human reader.

    Price: The client is free under Apache-2.0. Scans spend model usage on your ChatGPT plan or API key; the example GitHub Action sets an estimated $5 limit per scan, which in-flight requests can exceed.

    Read our Codex Security CLI and SDK review
    4.1/5
  7. 7

    Repowise gives a coding agent a map of the repository and gives the team a code health tool from the same index, at the cost of a slow first build and a large, fast-moving surface.

    Best for: Teams using Claude Code, Codex or Cursor on a large or unfamiliar repository who also want code health, change risk and test impact from the same local index.

    Price: Free to self-host under AGPL-3.0. Hosted repowise.dev: Free for 2 public repos; Pro $15/month for 5 private repos ($20 for new subscribers from 1 December); Teams $20/seat/month, minimum 3 seats; Enterprise custom.

    Read our Repowise review
    4.1/5
  8. 8

    safeagents adds a guard that works with any coding agent by sitting on the API line, and its design choices are sound, though it is a filter on known patterns, not a sandbox.

    Best for: Developers and teams running coding agents with real permissions who want a cheap, agent-neutral check on tool calls before they run.

    Price: Free and open source (Apache 2.0); an optional judge model adds its own cost.

    Read our safeagents review
    4.1/5
  9. 9

    Beads gives coding agents a task list that survives the end of a session and the merge of a branch, at the price of a database, frequent schema migrations and a set of upgrade traps you have to read about first.

    Best for: Developers running Claude Code, Codex or several agents on multi-day work who want the plan stored as queryable tasks rather than markdown.

    Price: Free and open source (MIT). Optional compaction and AI duplicate detection call a model on your own Anthropic or MiniMax key.

    Read our Beads review
    4.0/5
  10. 10

    The seatbelt for the damage a sandbox allows on purpose: force-pushes, hard resets and rm -rf inside your own project.

    Best for: Anyone letting a coding agent run shell commands without approving each one.

    Price: Free and open source under MIT.

    Read our CC Safety Net review
    4.0/5
  11. 11

    A capable local switchboard for running coding agents on cheaper models, as long as you accept that Anthropic will not support the result and the edits will be worse.

    Best for: Developers who want to run Claude Code or other agents on cheaper or local models.

    Price: Free and open source (MIT). You pay each model provider you route to.

    Read our Claude Code Router review
    4.0/5
  12. 12

    The sensible way to let Claude reason about a video today: local, cheap on tokens and honest about timestamps, though it shows the model stills and words rather than motion.

    Best for: Claude Code and other agent users who need a model to answer questions about videos without uploading them.

    Price: Free and open source (MIT). Optional crv Pro is a one-time $29; a separate Codex build is a one-time $9.90.

    Read our claude-real-video (crv) review
    4.0/5
  13. 13

    Headroom is the most rigorous of the token-saving add-ons, and its own numbers show the savings depend almost entirely on how repetitive your tool output is.

    Best for: Teams whose agents or RAG pipelines push large JSON, logs or search results into the model and pay per input token.

    Price: Free and open source (Apache 2.0). Headroom Enterprise for teams is sold separately; pricing not published.

    Read our Headroom review
    3.9/5
  14. 14

    Iris does one job for coding agents, taking a trustworthy picture of a page, and does it with less setup and less surface than a full browser-automation stack.

    Best for: Developers who want Claude Code, Codex or another MCP-capable agent to check its own front-end changes visually without wiring up a browser-automation framework.

    Price: Free and open source (MIT). Needs only a Chrome-family browser already installed.

    Read our Iris review
    3.9/5
  15. 15

    jevgrep is a well-measured way to cut a coding agent's file-hunting bill by about a quarter, provided you are content to send your source to a model provider for every search.

    Best for: Teams running Claude Code, Codex or OpenCode on unfamiliar repositories who already route model calls through a gateway and want to cut the cost of the agent's file search.

    Price: Free under MIT. Searches are billed by your Jev provider; Vercel AI Gateway lists Jev at $0.042 per million input tokens.

    Read our jevgrep review
    3.9/5
  16. 16

    Agent Autoloop is a carefully built way to let Claude Code or Codex work through a queue while you are away, but it is ten days old and its safety rests on the access level you choose, not on the hub.

    Best for: Developers who already run Claude Code or Codex and want scheduled or webhook-driven jobs to run unattended, with a phone notification for the decisions.

    Price: Free and open source (MIT). Model usage is billed to the Claude Code, Codex or other agent account you are already signed into.

    Read our Agent Autoloop review
    3.8/5
  17. 17

    Beacon keeps one clean record of every coding agent you use; the memory it builds is only as good as the lessons you stop to approve.

    Best for: Developers and platform teams who switch between several coding agents and want one auditable record of what they did, with memory they curate themselves.

    Price: Free and open source (MIT). Beacon Cloud is free with 50 findings; Pro $20/month ($15/month billed annually); Team $30/seat/month plus a $10/month base fee ($5 billed annually), up to 20 seats; Enterprise custom.

    Read our Beacon review
    3.8/5
  18. 18

    A tidy, very light shared task board for you and your agents, best kept to one machine or a trusted network until it grows up.

    Best for: Developers running Claude Code, Codex or other MCP agents who want one shared, local task board without adopting a hosted tracker.

    Price: Free under the AGPL-3.0 licence; no hosted plan.

    Read our Kanzo review
    3.8/5
  19. 19

    NotebookLM's full feature set in an agent's hands, held together by undocumented Google endpoints and a login you should not give it lightly.

    Best for: Researchers who want Claude Code to build and query NotebookLM notebooks.

    Price: Free and open source (MIT). NotebookLM limits depend on your Google account tier.

    Read our notebooklm-py review
    3.8/5
  20. 20

    oc cuts what a coding agent pays to read a web page by an order of magnitude on its author's benchmarks, but it only reads: no JavaScript pages and no forms yet.

    Best for: Developers whose coding agents read documentation, package registries and forums all day and who want to cut the tokens and cost of each lookup.

    Price: Free and open source under the MIT licence; no account, hosted service or paid tier.

    Read our only-cli (oc) review
    3.8/5
  21. 21

    A narrow fix for a real and costly problem, with an unusually careful benchmark behind it, but only for JavaScript and TypeScript and only from one maintainer.

    Best for: TypeScript developers whose coding agent keeps writing code for older versions of fast-moving packages such as Next.js, the AI SDK or zod.

    Price: Free under MIT. Optional --llm, --examples and --guides features call Claude on your own Anthropic key or Claude Code login.

    Read our stalemap review
    3.8/5
  22. 22

    Stroq is a sensible second line of defence for anyone whose agent reads the web, and its retroactive audit of past sessions is worth running once even if you never install the hooks, but it is a text filter from a small team and not a substitute for a sandbox.

    Best for: Developers who let Claude Code, Codex or Cursor read web pages, issues or cloned repositories and want a warning when the agent acts on what it read.

    Price: Free and open source (Apache-2.0). No account, no telemetry.

    Read our Stroq review
    3.8/5
  23. 23

    A cost cutter with a real study behind it, worth adding to repositories you work in every day, provided you expect less than the paper's headline from the shipped defaults.

    Best for: Developers and teams paying per token for Claude Code, Codex or Copilot CLI who run the agent in the same few repositories day after day.

    Price: Free under the Apache-2.0 licence; its own model calls are billed to your existing agent plan or key.

    Read our TokenEconomizer review
    3.8/5
  24. 24

    AgentFS turns an agent's files, state and tool history into one SQLite file you can query and copy, a clean idea that is still labelled beta and has gone quiet since the spring.

    Best for: Teams building agent platforms or sandboxes who want every agent run stored as one inspectable, copyable database.

    Price: Free and open source (MIT, per the README and package metadata). Optional sync goes to Turso Cloud, which is billed by Turso separately.

    Read our AgentFS review
    3.7/5
  25. 25

    agweb solves a narrow, real annoyance well, browser windows stealing your keyboard while agents work, but it is five days old, macOS 26 only and shares logins across agents by design.

    Best for: Mac developers on macOS 26 who run several coding agents against web apps and want to watch them without losing their keyboard.

    Price: Free and open source under the MIT licence.

    Read our agweb review
    3.7/5
  26. 26

    Real cross-session memory for Claude Code, wrapped in an installer that steers you towards a hosted subscription and a compression bill you have to watch.

    Best for: Developers on long-running projects who want the agent to remember past sessions.

    Price: Open source (Apache-2.0). Compression runs on your Anthropic plan or your own key. Hosted CMEM plans: Pro $30/month, Heavy $70/month, Max $100/month; automatic upgrades can move you up a tier, and Max bills overages; TeamBrain pricing not announced.

    Read our claude-mem review
    3.7/5
  27. 27

    A serious attempt to make unattended agent work trustworthy, with real evidence from its own board, though it is a month old and asks you to adopt its whole way of working.

    Best for: Developers running several coding agents on a long-lived product with a real test suite, who want agents checked by agents rather than reviewing every diff themselves.

    Price: Free under the MIT licence. The optional hosted relay has no published price.

    Read our Pullboard review
    3.7/5
  28. 28

    Caveman is a harmless way to shave a little off an agent bill, as long as you believe JetBrains' 8.5% rather than the repository's 65%.

    Best for: Developers paying per token for long agent sessions who want shorter replies and smaller tool output.

    Price: Free and open source (Apache 2.0). Auto needs a free Caveman account; Caveman Cloud is a separate hosted service.

    Read our Caveman review
    3.6/5
  29. 29

    Worth trying for long sessions that lose the thread after compaction; nobody has yet shown the free plugin lowers your bill.

    Best for: Claude Code users whose long sessions lose track of their work after compaction.

    Price: Plugin free (Elastic License 2.0). Hosted gateway: 1,000 free requests once; Starter $20/month (10,000 requests); Pro $49/month (50,000 requests); you pay your model provider directly.

    Read our context-mode review
    3.6/5
  30. 30

    An unusually capable inspection engine wrapped around a rebuild loop that is still early alpha, and one that puts the legal judgement squarely on you.

    Best for: Developers and reverse engineers who want an MCP-accessible inspection engine and a Claude Code rebuild loop for prototypes and study, accepting alpha-quality software.

    Price: Free under MIT or Apache-2.0. You pay for your coding agent (such as a Claude subscription) and optional fal.ai image generation; a managed tier is planned but not priced.

    Read our Craft Anything review
    3.6/5
  31. 31

    Seeing where an agent wandered in your repository takes seconds with mindwalk, but it is a young tool whose development has gone quiet since August.

    Best for: Developers and tech leads who review coding-agent work and want to see an agent's footprint before reading its diff.

    Price: Free and open source (MIT). Session evaluation uses your existing Claude or Codex plan.

    Read our mindwalk review
    3.6/5
  32. 32

    Pinky has the most careful safety design of any agent phone tool we have read about, but it is a week old, and its own FAQ admits an agent with a shell can walk around it.

    Best for: Developers who want an MCP agent to work through apps on a spare or test phone, with a person approving the steps that spend money or send messages.

    Price: Free under Apache-2.0 with no account. The built-in agent uses your own model API key; its default task limit is US$2.

    Read our Pinky review
    3.6/5
  33. 33

    A clever answer to the subscription wall around agent data, if you are comfortable handing a young service both your spend and your API keys.

    Best for: Growth, sales and research teams whose agents need occasional calls to paid data APIs they would not otherwise subscribe to.

    Price: Self-hosting is free under a modified Apache-2.0 licence. Hosted treg.to charges per call from a prepaid balance (examples: $0.0018 per Semrush keyword, $0.01 per Crunchbase call), with $1.00 of free credit once per verified account; calls on your own keys are not metered.

    Read our treg review
    3.6/5
  34. 34

    A clever answer to agents running out of context, delivered as a self-updating, unauthenticated proxy that sees every prompt and key you send, so it asks for more trust than most teams should give it.

    Best for: Individual developers on their own machine who run very long coding-agent sessions and are prepared to pin versions and review what the proxy does.

    Price: Free and open source (MIT with an added attribution term). You still pay your model provider for every token the proxy forwards.

    Read our billion-context review
    3.4/5
  35. 35

    It makes shell output shorter, which is not the same as making Claude Code cheaper, and both independent cost tests found it was not.

    Best for: Developers with very noisy shell output who will measure the effect themselves.

    Price: Free (Apache-2.0 CLI plus BUSL-1.1 engine; Apache-only rtk-oss available). Commercial resale needs a licence. RTK Pro team edition is paid, with a 7-day trial.

    Read our rtk (Rust Token Killer) review
    3.4/5
  36. 36

    BLA is a clever, unusually candid answer to one specific tool, REA, and your release builds pay for it in megabytes while a determined human analyst can still get through.

    Best for: Developers shipping Electron, JavaScript or WebAssembly products who want to make automated agent reverse engineering more expensive and can accept larger builds.

    Price: Free under the MIT licence. No paid tier; optional Wabt and tshark are only needed to rerun the red-team checks.

    Read our BLA (Blind LLM Agents) review
    3.3/5
  37. 37

    Mantic is a quick, private way to point an agent at the right file when your paths are well named, and little help when they are not.

    Best for: Developers whose agents work in well-structured TypeScript or Python repos and who want a free, local first-pass file ranker.

    Price: Free and open source (MIT from v1.0.26; earlier releases AGPL-3.0). Runs locally with no paid tier.

    Read our Mantic.sh review
    3.3/5
  38. 38

    REM has a sound method for reading a rival's marketing from public traces, wrapped in a one-day code drop whose advertised install command does not work.

    Best for: Growth marketers and founders who can run a Node project and want a cited first read of a competitor's channels before paying for a research tool.

    Price: Free under MIT. The free sources need no key. Ad libraries, social and Reddit data use Scrape Creators at about 25 to 50 credits a run (100 free credits; packs from $47 for 25,000). The report runs on your Claude or ChatGPT plan, or on the Anthropic API per token.

    Read our REM (Reverse Engineer Marketing) review
    3.3/5
  39. 39

    Agent Reach is a clever convenience for personal research on Chinese and social platforms, and a liability anywhere an employer, a platform's terms or your main account matter.

    Best for: Individual researchers and developers who need their agent to read Chinese platforms and accept the account and terms-of-service risk on a secondary account.

    Price: Free under MIT. Some channels need your own keys: YouTube transcription uses a Groq or OpenAI key, podcast transcription a Groq key. The newest CloakBrowser binary needs a paid CloakHQ subscription.

    Read our Agent Reach review
    3.2/5
  40. 40

    A well-reasoned agent team for opencode users that enforces roles by permission rather than by polite prompting, but small, quiet since August and tied to one host.

    Best for: opencode users who want a ready-made multi-role agent team with hard permission boundaries and are comfortable with a young, single-maintainer project.

    Price: Free under the MIT licence; you pay for the models opencode calls.

    Read our armada review
    3.2/5
  41. 41

    social-auto-upload reaches more Chinese and Western platforms than any open tool we know, by doing exactly the kind of automated posting those platforms ban accounts for.

    Best for: Individual creators and small studios cross-posting their own videos to Chinese platforms who accept the account risk of browser automation.

    Price: Free under MIT. No platform fees; you need your own accounts on each site and, where a site is blocked in your region, a proxy.

    Read our social-auto-upload review
    3.2/5

Sources

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

More guides