Open Source Workflow Automation
As of August 2026, AIDiveForge tracks 20 open source workflow automation. The top three by verified-data score are Desktop Commander MCP, MobileRun, and ego-lite. Curated open source workflow automation tracked by AIDiveForge. Each project has a verified public source repository. Listings are verified against each tool's live website and re-checked regularly.
Last updated July 29, 2026 · 20 tools
Ranked by AIDiveForge's verified-data score: data completeness, verification recency, community rating, and real visitor engagement. How we rank · No tool can pay for placement.
1. Desktop Commander MCP
The app runs locally on your machine, reads and writes your files directly without uploads, and exposes every action it takes in plain view as it works. You describe the outcome in plain English; the agent figures out the steps across your filesystem and connected apps. The model roster is provider-agnostic — Opus, GPT, Gemini, or whatever fits — so you are not locked to one vendor's API pricing. Where it strains: teams needing headless, server-side automation or multi-user pipelines will hit the single-machine ceiling fast. At that point, teams move the logic into a backend orchestration layer and treat Desktop Commander as a local prototyping step they've outgrown.
PaidOpen Sourcefrom $20/monthSelf-hostedVerified Jul 13, 2026
2. MobileRun
The platform pairs an LLM agent — Claude, Hermes, or their own Mobilerun VA — with a persistent real device that has a real SIM, a real residential IP, and a real number receiving SMS. The agent reads screen state via accessibility tree plus vision, then taps, swipes, and types end-to-end. That identity stack is why apps like Instagram, WhatsApp, and Stripe behave the way they do in production rather than flagging the session. Fleet tooling handles parallel runs across up to 10,000 devices. The ceiling appears when workflows require branch logic the natural-language task description cannot express — at that point you are writing SDK code.
PaidOpen Source$5/monthAPISelf-hostedVerified Jul 9, 2026
3. ego-lite
ego (lite) is a custom Chromium build that installs as your daily browser and exposes a skill called ego-browser, which any code-writing agent — Claude Code, Codex, Cursor, Kiro — can drive directly. Agents run inside isolated Spaces so they don't collide with your open tabs. The vendor states the engine handles cross-origin iframes, shadow DOM, and third-party widgets like Stripe and Salesforce that JS shims typically fail on. The page claims task completion up to 3.45x faster than agent-browser tooling, on fewer tokens, because multiple in-page actions batch into a few lines of JavaScript instead of one tool call at a time. Mac-only at this point — Windows support is on a waitlist.
FreeOpen SourceAPISelf-hostedVerified Jul 28, 2026
4. Skillful
The core mechanic is symlink-based installation: one source folder, linked into Claude Code, Codex, Cursor, Copilot, Gemini CLI, Junie, opencode, or any custom folder target you define. Edit the source and every linked install picks up the change. The docs describe broken-symlink detection and one-click repair, which matters when tools update their config paths and silently drop your workflows. There is no backend, no account, and no telemetry — the library is a folder on your disk, so git is your version control and your backup strategy. Public skill repositories on GitHub can open directly into Skillful's import dialog, which makes trialing shared libraries faster than copying URLs by hand.
FreeOpen SourceSelf-hostedVerified Jul 18, 2026
5. Wayflow
The vendor describes Wayflow as an embeddable visual workflow editor — canvas, node palette, config panel, run controls, and a runtime engine that executes the same graph in the browser during development and on your server in production. A single call, createWorkflowEditor(), mounts the full workspace. Built-in node types cover LLM calls, tool-calling, branching, map-over-list, and image generation, so AI and deterministic steps sit on the same canvas. Suspend-and-resume is native, meaning a workflow can pause for a human review and pick back up without custom state management. The project is MIT licensed, carries zero runtime dependencies according to the docs, and ships full TypeScript types.
FreeOpen SourceAPISelf-hostedVerified Jul 8, 2026
6. Browser-Use
The open-source library installs via pip and gives an LLM-controlled browser the ability to complete multi-step tasks — form filling, navigation, data extraction — with a self-healing loop that recovers from DOM changes without manual patching. The cloud tier adds anti-detect browsers, CAPTCHA solving, and residential proxies across 195+ countries, all zero-config. The vendor reports 97% benchmark accuracy on browser tasks, and cold start times are described as sub-second. At production scale, the hosted agent layer handles parallel sessions, so the constraint shifts from infrastructure to prompt quality and LLM cost. Teams running high-frequency extraction at scale will hit the point where LLM call costs per task matter more than the browser infrastructure fee.
PaidOpen Source$29/moAPISelf-hostedVerified Jun 30, 2026
7. BrowserAct
The core loop is prompt-in, structured-data-out: describe what you need, the agent builds and tests a Bot, then publishes it as a reusable scraper you can trigger from Make, n8n, or Zapier. Built-in residential proxies and CAPTCHA handling mean protected pages are reachable without assembling your own infrastructure. The local agent option lets teams run browsers on their own hardware when data cannot leave the building. The ceiling appears when your extraction logic requires conditional branching across multiple page types — the prompt interface has no canvas for that, so complex workflows still need a surrounding orchestration layer. Community ratings on G2 sit at 4.6, suggesting the core promise holds for straightforward collection tasks.
PaidOpen SourceFree Trial · 7 daysAPISelf-hostedVerified Jul 29, 2026
8. OfficeCLI
The tool ships as a self-contained binary with a skill-file interface, so an agent can create, edit, and analyze Office documents by invoking one-line commands — no GUI, no COM interop, no Office install on the host. That headless design is the point: it targets CI pipelines, container environments, and coding agents that need document automation as a side effect of a larger workflow. Where it starts to strain is when document fidelity gets complex — deeply nested styles, advanced Excel formulas, or PowerPoint animations that depend on Office's own rendering engine may not survive the round-trip. Community reports on the GitHub issue tracker flag edge cases in format preservation. Teams hitting those limits typically fall back to Office Open XML manipulation libraries directly.
FreeOpen SourceSelf-hostedVerified Jul 14, 2026
9. OpenIngress
OpenIngress is an open-source, self-hosted tool that scans public-facing websites and reports where AI agents will break: unlabeled interactive elements, flow blockers before key actions, and UI patterns that confuse automated navigation. The core workflow is pointed at a URL — the tool crawls the target and surfaces specific compatibility issues along with fix recommendations aimed at developers. It fits cleanly into pre-deployment audits for e-commerce flows, form submissions, or any checkout-critical path. The self-hosted model means scan results stay on your infrastructure. The scrape-and-report approach covers what's publicly visible, which is the ceiling — authenticated flows and dynamic states behind login require additional tooling.
FreeOpen SourceAPISelf-hostedVerified Jul 23, 2026
10. Qpilot
You paste a test case written the way you'd explain it to a colleague, run a single npx command, and the agent opens Chrome, navigates, clicks, and fills forms step by step. Each step returns a pass, fail, or warn in real time, with screenshots captured on failure so you don't have to reproduce it manually. The agent pauses and prompts you when it hits an OTP or anything requiring live credentials, then continues — so you stay in the loop without babysitting the whole run. The ceiling appears fast on headless CI environments and parallel test execution, because the architecture requires a local Chrome instance and an Anthropic API key, both of which add cost and coordination overhead at scale.
FreeOpen SourceSelf-hostedVerified Jul 6, 2026
11. Tsaagan
The architecture centers on perception-action-verification loops rather than fire-and-forget scripting, which means each browser action waits for confirmed state before the agent proceeds. Tsaagan ships an MCP server alongside JS and Python SDKs, so agents already wired into those runtimes can call browser actions without building a separate automation layer. It runs on Playwright, native APIs, and a browser extension — giving it reach across sites that block headless fingerprints. The public repo shows 29 commits and three open issues, which signals early-stage software; production teams should expect rough edges and plan to contribute fixes. For simple, authenticated scraping pipelines it earns its place — for high-volume, concurrent agent fleets the maturity ceiling appears quickly.
FreeOpen SourceAPISelf-hostedVerified Jul 22, 2026
12. ComfyUI
ComfyUI exposes every model, sampler, conditioning step, and post-processing node as a discrete block on an infinite canvas, so the pipeline that produced the result is the documentation. VFX teams and creative studios use it to build multi-model pipelines where every parameter is auditable and every step can be swapped independently. The node graph runs locally via Comfy Desktop or in the cloud, and workflows can be published as API endpoints through Comfy API. Where it breaks: the node graph is not a gentle on-ramp. Teams that need non-technical colleagues to run pipelines use App Mode, a simplified overlay, but deep customization still requires fluency with the underlying graph.
PaidOpen SourceAPISelf-hostedVerified Jun 30, 2026
13. AI Mime
AI Mime records a macOS task once, then compiles the raw trace into a coordinate-free skill: deterministic scripts where possible, a browser harness or native UI agent only at decision points where necessary. The self-healing loop is the real differentiator — when a run fails, an agent reads the logs, triages the issue, and patches the skill instead of silently dying. The output is a readable directory of files, not a locked binary, so Claude Code or Codex can call it directly. The wall appears on Windows and Linux: this is macOS-only, and teams needing cross-platform coverage will hit that ceiling before the third workflow.
FreeOpen SourceSelf-hostedVerified Jun 12, 2026
14. chrome-use
Because chrome-use attaches to the Chrome instance you already use day-to-day, the agent inherits your cookies, your fingerprint, and your authenticated state — no credential injection, no headless flags that anti-bot systems fingerprint in milliseconds. The extension-relay architecture means the browser looks identical to your normal traffic. Where it breaks: tasks that need a clean isolated session per run, or parallel agents that must operate on separate authenticated identities simultaneously, get complicated fast because you are sharing one browser. Teams hitting that ceiling wire in additional Chrome profiles or move the workflow to a cloud browser infrastructure that supports session isolation at scale.
FreeOpen SourceSelf-hostedVerified Jun 25, 2026
15. cua
Cua provisions cross-OS fleets from a single API, forks machine state over copy-on-write snapshots so you can reproduce failures without rebuilding from scratch, and serves pre-booted machines from warm pools that claim in milliseconds. The open-source Cua Driver runs background desktop automation on macOS and Windows — agents click, type, scroll, and inspect accessibility trees without stealing your cursor. Linux support in Cua Driver is in pre-release, so teams with Linux-heavy desktop workflows will hit that wall immediately. At scale, you either point your training loop at live warm pools or order verified trajectory datasets that arrive pre-packaged for your ingestion pipeline.
PaidOpen SourceAPISelf-hostedVerified Jun 18, 2026
16. GhostUser
Each persona — a cautious newcomer, a skeptical evaluator, a power user, a time-pressured visitor, a motivated buyer — navigates your app autonomously, flags where it gave up, and logs why. Console errors, failed network requests, and 5xx responses get caught in the same pass, so you get UX feedback and QA signal in one run. It connects directly to localhost, which means you catch issues before they leave your machine. The tool runs on your Claude API key, so cost scales with usage rather than with a seat count. Where it breaks: the feedback reflects what five hardcoded personas notice, not the distribution of your actual users.
FreeOpen SourceFree (pay-per-token for Claude API calls)APISelf-hostedVerified Jun 9, 2026
17. Peerd
peerd is a browser extension that turns your existing browser into an agent workstation: the agent shares your tabs, your authenticated sessions, and your stored credentials without any cloud relay, background process, or external tool broker. It runs Linux VMs via WebAssembly, executes JavaScript notebooks, and connects browser agents peer-to-peer over WebRTC — all client-side. The architecture is genuinely serverless in the literal sense: there is no server. That zero-server posture is also the ceiling: any workflow that needs a persistent, always-on agent, a team-shared backend, or centralized audit logs runs into a wall. Teams needing those properties will need to wire up their own coordination layer or move to a hosted platform.
FreeOpen SourceSelf-hostedVerified Jun 24, 2026
18. RiddleRun
RiddleRun combines a CLI and an optional self-hosted web app, both running inside Docker, so your test environment travels with the repo rather than living on someone's laptop. You define a user journey in JSON — steps, assertions, expected outcomes — and a Playwright/browser-use agent executes the whole sequence autonomously. The Docker-first setup means teams can wire it into CI without installing a browser stack on the build machine. The project has two GitHub stars and one open issue at the time of curation, which signals early-stage maturity — documentation depth and community support are thin, and the agent's decision logic is largely a black box to the teams running it.
FreeOpen SourceSelf-hostedVerified Jun 11, 2026
19. SoMatic
The core workflow is a CLI command that takes a screenshot, runs element detection locally, and returns numbered marks with coordinates as JSON — so agents target elements by ID, not by fragile pixel hunts. Every action returns JSON, which means downstream agents can chain steps without parsing unstructured output. The self-hosted, MIT-licensed model runs on your own hardware, so no screenshot data leaves the machine. The wall appears with non-standard or highly dynamic UIs where YOLO detection misses elements or mislabels them — teams handling those cases add a fallback coordinate layer manually. At this GitHub star count, the community size is small, which means debugging edge cases happens in the codebase, not a forum.
FreeOpen SourceAPISelf-hostedVerified Jun 9, 2026
20. Vessel Browser
Vessel exposes 80+ browser tools over MCP — navigate, click, fill, extract, diff, checkpoint, undo — so any MCP-compatible agent harness can drive a real browser without brittle Playwright scripts or headless guesswork. The visible-by-default design means you can watch, pause, approve, or roll back before something ships. Session persistence and named checkpoints mean the agent picks up where it left off instead of re-authenticating from scratch every run. The wall appears on complex, multi-site workflows that branch based on what the last page returned — the built-in skill system handles linear recipes well, but conditional logic across domains requires more than reusable instructions. Saved sessions, credential vaults, and page-change history are paid-only features, so teams running fully autonomous monitoring pipelines hit that ceiling fast.
PaidOpen SourceFree Trial · 7 days$5.99/monthAPISelf-hostedVerified Jun 19, 2026
Listings on this page are sourced and verified by the AIDiveForge data pipeline. AIDiveForge is editorially independent — no money changes hands for inclusion.