Overview
Flowith calls itself an agentic workspace rather than a chat box. Tasks sit as nodes on a canvas, agents run on those nodes, and every branch, merge and rollback leaves a trace you can inspect. The form has been in beta since 2024, has cycled through GPT, Claude and Gemini models, and now ships its own Agent Neo and Deep Research.
The problem it solves is specific. When a task goes sideways halfway through, a chat interface only offers starting over. On a canvas you can detach that branch, keep whatever worked alongside it, and continue. Nodes run in parallel, so three approaches to the same job can be attempted at once before you pick one.
Four surfaces make up the product. Agent Neo is the standing executor that takes a task, breaks it into steps, calls tools and produces files. Deep Research handles multi-source retrieval and drafting. Slides and Website turn results into a deck or a live page. Over 40 models sit underneath, mixing text, image and video, with the Nano Banana and Seedream families on the image side and Seedance and Kling on video.
Key Features
- Canvas-based orchestration: Tasks lay out as nodes on an infinite canvas. Branches run in parallel and merge on demand, and a wrong branch can be detached without disturbing what already worked. Intermediate artefacts stay visible.
- Agent Neo as executor: Give it one sentence and it plans steps, calls tools, searches the web and returns documents, sheets or pages. You can inject new instructions mid-run instead of restarting.
- Deep Research to a finished draft: It runs several retrieval rounds on a topic and assembles sources, arguments and conclusions into a cited long-form piece, with intermediate drafts kept in nodes for section-level rewriting.
- Slides and Website output: Research can be turned into a deck or a shareable page in one step, with layout and imagery generated, and you can still return to the canvas to restructure before regenerating.
- More than 40 models: Text, image and video models coexist in one workspace, and different nodes in the same task can specify different models to balance cost against quality.
- Recipes for reuse: Save a working node sequence and its parameters as a Recipe, then rerun it with new inputs for recurring reports, topic selection or batch asset production.
Use Cases
- Researchers laying out three argument paths at once and keeping whichever one survives contact with the sources.
- Content teams running topic, sourcing, imagery and drafting as four parallel nodes that merge at the end.
- Anyone producing the same report weekly, saving the flow as a Recipe and swapping the data source.
- Product planning that starts with Deep Research on competitors and pricing, then becomes a deck on the same canvas.
- Creators calling image and video models in one place instead of switching between five sites.
- People who want to see what an agent is doing, since the canvas exposes each step for intervention.
Pros
- Task structure is visible, with branches and merges leaving a reviewable trail.
- Trying several paths in parallel is cheap, since a wrong branch can simply be detached.
- Agent Neo and Deep Research both deliver finished artefacts rather than a paragraph of answer.
- Model choice is broad, and text, image and video work happen in one workspace.
- Recipes turn repetitive work into a one-time setup.
- A free tier exists and Pro sits in the twenty-dollar range, so entry is cheap.
Pricing
Starter is free with 300 one-time credits and limited model access. Pro costs 19.90 dollars a month, or 17.91 billed yearly, and includes 20,000 credits with monthly allowances for Agent Neo around 36 runs, Slides around 22 and Website around 92. Ultimate is 49.90 dollars a month with 50,000 credits plus the unlimited pack, opening top-tier image and video models on day-based terms. Infinite is 499.90 dollars a month with 500,000 credits and unlimited concurrent tasks, with API access and custom integrations listed as coming. Credits deduct at per-model rates, and image or video generation consumes separately per run.
Summary
Flowith bets on agent visibility. Most agents hand you an input box and a result, leaving the middle opaque; Flowith lays the middle out as nodes so each step can be read, edited and rerun. That shape suits research, proposals and batch asset work, because it lets you change your mind mid-task rather than discovering the wrong direction after a long run finishes. Forty-plus models and three output modules mean one workspace covers the path from retrieval to finished file.
Version History
- Flowith added with its canvas agentic workspace (2026-10-29): As listed: an infinite canvas for orchestrating task nodes with parallel branches and merges; Agent Neo for standing execution, Deep Research for cited drafting, and Slides plus Website for direct output; over 40 text, image and video models with Recipes to save working flows as templates. Pricing runs from a free Starter tier with 300 one-time credits to Pro at 19.90 dollars, Ultimate at 49.90 and Infinite at 499.90 a month.