What Samy does on a workflow
Samy works inside the graph: it adds, connects, edits and deletes nodes, asks a question when the brief is unclear, proposes a plan, launches a run or a batch, and tidies the canvas so nodes keep from overlapping. It relies on ten tested recipes (brief to image, reference to three formats, reference to video, brief to image then video…) and can build a four-shot storyboard around one character reference, with no guarantee of perfect consistency from shot to shot.
3 modes
Ask, Assisted, Auto: you choose how far it goes
10
tested workflow recipes, from a single image to a storyboard
7
specialists who advise and never apply changes on their own
How Samy works, in four moves
Move 1
You describe the need
In the workflow chat panel, typed or spoken: dictation transcribes up to 60 seconds per message, the text lands in your draft and is never sent on its own. Dictation costs no credits.
Move 2
Samy reads your context
It looks at the Samy Brain spaces you turned on and keeps only the excerpts relevant to the request, plus the active skills. When the brief is missing something, it asks rather than guesses.
Move 3
It builds or proposes
In Ask mode it answers without changing or generating anything. In Assisted mode it proposes a plan for you to approve. In Auto mode it applies changes directly. Shift-Tab switches mode. Generations, which spend credits, wait for your confirmation, even in Auto.
Move 4
You stay in control
Every request to Samy sets a checkpoint: you go back to the canvas as it was before its answer, except while a run is in progress, and the rollback applies to the shared canvas. A run that stops without a confirmed result is flagged as interrupted, never as a success.
The building blocks of Samy
| Block | What it is | What you control |
|---|---|---|
| Modes | Ask, Assisted, Auto | The level of autonomy, per message |
| The Samy Brain | Your notes and brand guide, in personal spaces | Which spaces Samy reads; the scope of each note (Global, Project, Brand) |
| Skills | Methods: 11 built in, 10 Beemm Methods, your own | Which methods are on; a skill grants no tool |
| Specialists | Seven roles called by @ mention | Every proposal waits for your confirmation |
| Sub-agents | Parallel prompt rewriting, feedback on the visuals | Feedback shows in the chat; nothing is fixed without you |
| Checkpoints | One way back to the canvas before each request | Not during a run; no global rollback after a failure |
The Samy Brain, the skills and the specialists
The Samy Brain
Personal note spaces, each on or off for Samy, and notes filed by scope: Global (your preferences), Project (this brief, shared with the project), Brand (the shared rules). Block notes with images, video and logos, links between notes, and a brand guide imported from a website: colours, logo, images, which you approve before the notes are created.
Skills, not powers
A skill brings a method, never an extra tool. Eleven are built in (image analysis, model choice, prompt writing, graph building…), ten Beemm Methods turn on when you need them (Film Director, E-commerce Packshot, UGC Creator, Pro Portraitist, Character Designer…), and you write your own, kept private. Community skills are published after moderation.
Seven specialists who advise
Architect, Prompt writer, Script supervisor, Director, Art director, Producer and Choreographer. You call them by mention; they work from the text of the graph and the brief, and their proposed changes wait for your approval. They do not trigger each other as a team.
Bounded sub-agents
To move fast, Samy hands out narrow tasks: one sub-agent rewrites the prompts of several nodes in parallel, another can compare the generated visuals with the brief and the brand guide and write feedback in the chat. Samy alone keeps the graph and the conversation.
Frequently asked questions
- Is Samy available to every account?
- Not yet: Samy is in testing and will be available to everyone soon. Meanwhile, the ✨ sparkle on a personal-project workflow lets you drive it with Claude or ChatGPT.
- Can Samy change my workflow without my approval?
- In Ask mode, no: it answers without changing anything. In Assisted mode, it proposes a plan for you to approve. In Auto mode, it applies changes directly. Generations, which spend credits, and specialist proposals always wait for your confirmation.
- What is the Samy Brain?
- Your personal context spaces: block notes, a brand guide you can import from a website, links between notes. You choose which spaces Samy reads, and it only takes the excerpts relevant to the request.
- Can I teach it my own methods?
- Yes, with “My skills”: private instructions, active right away. A skill guides Samy; it does not give it any extra tool.
- Are my conversations with Samy private?
- Yes: conversations, runs and the automatic memory are private to their author. Notes with the Project scope are shared with the project, and the canvas stays shared.
- How much does Samy cost?
- Samy spends credits on each request, depending on the model you pick and the work done; the model menu shows the prices. The generations it launches are billed separately, after your confirmation. Dictation costs nothing. A request that fails for a technical reason is refunded; if you stop it mid-way, only what was already used stays charged.
Until Samy opens: drive your workflow with your own AI
Create your account, open a workflow in a personal project, then click the ✨ sparkle: “Lancer avec ChatGPT” (launch with ChatGPT) or, on a computer, “Ouvrir dans Claude à la place” (open in Claude instead); Claude Code connects read-only for now. The app’s buttons are in French. Your AI reasons on its own subscription; only generations spend credits.