Manus alternatives for agents and browser tasks

DeerFlow runs research, code and content agents with browser, shell and file tools in Docker sandboxes. Skills produce reports, slides, webpages and data visualizations. It suits Manus users who want a broad task workspace they operate on their own infrastructure.

Goose suits tasks that combine coding, research and connected tools. Its desktop app and CLI can use Ollama or hosted providers, with MCP extensions for outside services.

nanobot covers tasks that combine files, shell commands and web research. Its browser workbench shows tool calls, diffs and generated artifacts, while memory, schedules and messaging channels support ongoing work. Scheduled delivery needs the gateway running.

LobeHub suits delegating research and writing to several agents in a self-hosted workspace. Agent groups share project context and can schedule work, with skills and MCP tools extending what each agent can do.

Agenta suits teams sharing agents for document work, website research and recurring tasks. Agents work in shared file spaces and can run on schedules or connected-app events, with permissions and approval rules set by the team.

LobsterAI suits office tasks that need finished files: spreadsheet analyses, Word documents, presentations or PDF processing. Its desktop interface shows progress and previews the outputs, while scheduled jobs can handle recurring briefings.

OpenHuman turns a task into an inspectable workflow that can involve several agents and connected apps. Saved workflows can resume after a pause or start from a schedule or app event; persistent memory requires a connected CortexDB engine.

WorkAny handles file organization and creates documents, spreadsheets or presentations through a desktop agent. Artifact previews let you inspect generated work, while its shell and web tools support tasks that need several steps.

Row-Bot covers personal research, editable documents and scheduled workflows in one workspace. Focused agents can share longer tasks, while a knowledge graph and inspectable conversations retain project context. The server has one owner rather than separate team users.

HanaAgent covers file, code and browser tasks through a desktop agent with persistent memory. Agents can delegate to one another and continue scheduled work, with separate workspaces for their files and notes.

Enconvo works beside Mac apps with screen or selected-text context for browser and computer tasks. Its visual workflow editor combines tools, prompts and conditions for repeatable jobs, with MCP and skills for additional actions.

Hermes Agent suits recurring personal tasks that need memory across conversations. Its scheduler runs jobs such as reports and audits, while messaging channels let you continue the same work and reusable skills preserve what the agent learns.

gptme handles files, Python analysis and browser or desktop tasks in your own working environment. Persistent agents keep a Git-tracked journal and task queue, with schedules and child agents for work that continues across sessions.

Atomic Agent can read documents, organize files and operate a browser to research a task. It saves cited notes in searchable memory, and successful procedures can become reusable skills for later work.

OpenAgent gives a personal or team assistant browser access, code execution and tools for working with Office files. You can build workflows with branches and parallel steps, schedule them and follow the agent's visible tool calls.

Hope Agent carries goals across sessions with visible progress, budgets and pause/resume controls. With permission, it can work in the browser or desktop, edit files and run commands, then retain the task's notes for later work.

QM suits teams working with agents in Slack or shared browser projects. Each person and room has its own memory, files and permissions; agents can search workplace sources or schedule follow-ups. Configure persistent storage to retain sessions after a restart.

Odysseus gives agents shell and file tools alongside reusable skills and persistent conversation memory. The workspace also handles research, email triage and scheduled tasks, so a job can involve both an answer and actions.

PyGPT brings project document indexes and task agents into one desktop workspace. Agents can execute Python, use files and APIs, or run scheduled jobs, with an orchestrator for coordinating work and configured safeguards for code execution.

Lemon AI researches topics, processes documents and runs code in a Docker sandbox. Its report workspace lets you revise selected HTML sections or edit the text directly, with conversation memory for returning to previous work.

Skales keeps working on goals after you close the chat and can resume them after a restart. Plans and approval controls govern its actions, while paired devices let you delegate work or check a desktop task from the mobile app.

Open Cowork suits delegated document work and desktop app tasks. Its built-in skills handle Office files and PDFs, and a trace panel shows the agent's actions. The workspace executes tasks on your computer but uses a cloud model API.

UI-TARS Desktop covers tasks in graphical apps through screenshots, clicks and typing. Its interface shows progress while the agent works; choose whether it controls your computer or uses a remote operator environment.

Skyvern covers repeat browser work such as filling forms, navigating vendor portals and collecting structured data. Create a workflow from instructions or a browser recording, then inspect its actions and download results as JSON or CSV.

WebBrain handles page research and form tasks in the browser tabs you already use. Successful runs become reusable workflows, with schedules or page conditions that trigger them. Permission prompts govern consequential actions.

Nanobrowser delegates repetitive web tasks and research to planning and browsing agents inside Chrome or Edge. It suits Manus browser work through a ready extension, with Ollama or provider APIs processing the task.

AnythingLLM suits agent tasks grounded in your own documents. Agents can use workspace knowledge, browse the web and run scheduled tasks, with a desktop app or a shared Docker installation.

OpenHands suits developers who want a coding agent platform on their own machine or server. It provides a broader agent environment rather than a ready-made hosted app builder.

Agency Orchestrator suits research or content projects that benefit from several specialists. Arrange their work on a visual canvas, inspect progress and revise the resulting reports; acceptance checks and human sign-off can guide delivery.

Open Science Desktop suits research projects that need experiments as well as a literature review. Agents produce notebooks, figures and manuscripts, with outputs linked to the code and inputs behind them. It is a scientific workbench rather than a general task assistant.

Dive lets a model work with local files, shell commands and connected MCP tools from a desktop app. You choose which tools it can use and can add servers through its installer agent; capabilities depend on the tools and model you configure.

NullClaw runs agents that edit files, use Git and search the web under configured approval policies. Persistent sessions and scheduled tasks support background work, with terminal access and separate browser-dashboard tools for monitoring and approvals.

Browser Use suits developers automating multi-step website tasks through a Python agent or CLI. You choose the model and browser; managed remote browsers remain optional.