Skip to main content

Your AI workflow, visualized

A node-based visual canvas where you drag, connect, and orchestrate AI models into workflows that do real work. See the shape of your thinking, not just the output.

Memorandai's Flows Canvas in action — a wide overview of a working project with roughly 15 connected nodes spread across the canvas. The top toolbar shows the project selector, a Canvas/Classic mode toggle, and five workflow-mode sliders (Work Mode, Node Menu, Model Selection, Follow Mode, Auto-Memo). Below it: a Provider Models row with active model selections for each cloud provider (GPT-5.5, Claude Opus 4.7 1m, Gemini 3.1 Pro, Grok 3) plus the local Qwen3 model. A node-creation toolbar offers Basic / Advanced / Agent / Custom / Templates categories with quick-add buttons for each provider (Local, Claude, Gemini, GPT, Grok, Helios) and structural nodes (Roundtable, Memo, Text Output). The canvas itself shows a complex graph of color-coded chat, context, and output nodes connected by amber, green, blue, purple, and red edges in a roughly horizontal flow. A minimap in the bottom-right gives a navigation overview of the full graph.

The Flows Canvas is your primary Workspace in Memorandai. Whether you're importing conversations from other AI platforms, or starting fresh, this is where the bulk of your conversations and work with AI will take place. In the Flows canvas, you'll attach, branch, and chain different nodes to engage with AI in a creative fashion that doesn't need to be strictly linear.

Multi-model freedom

Use Claude, GPT, Gemini, Grok, and local models side by side on one canvas. Switch providers per-node, compare outputs, chain different models together. Your workflow, your choice of intelligence at every step.

A four-node chain on Memorandai's Flows Canvas showing the same conversation thread routed through different AI models in sequence. Each chat node has its provider's signature accent color across the top, a content area showing the model's response to the upstream context, and a footer with character/token counter and Send button. Amber edges connect each node to the next — illustrating per-node provider switching where the conversation continues across providers without losing the upstream context built up before this point.
  • Per-node provider selection — no global lock-in
  • Run local GGUF models with native GPU acceleration
  • Bring your own API keys — your accounts, your usage

Roundtable

Send one question to multiple AI models simultaneously. Each model responds individually, then output assembles in two parts -- a mechanical Roundtable Synthesis showing where models agree and disagree, followed by a Helios Perspective that applies your identity lens. If tool calling is enabled, models can call tools like web-search or memory-search during the Roundtable to ground their responses.

A live Roundtable example on Memorandai's Flows Canvas. The Roundtable input node on the left holds the user's question and a row of configuration controls — Participating Models checkboxes (Claude, GPT, Gemini, Grok, Local), a Rounds selector, a Synthesis model dropdown, a Tool Use toggle, and Cancel / Run Roundtable buttons. The Roundtable Response node on the right shows the multi-section output: a Roundtable Synthesis with Brief Overview, Points of Agreement, Points of Disagreement and Divergence, and Major Considerations sections, followed by a Helios Perspective section that re-reads the synthesis through the user's identity layer.

Tool calls in plain sight

When a model says it searched the web, checked your memory, or pulled from a file, you shouldn't have to take its word for it. Every tool call surfaces inline as a collapsed row that you can expand to see the exact arguments sent and the exact response returned. A green checkmark per call confirms it executed; the panel header tallies how many ran.

It's a small thing that changes how much you can trust a longer answer. If a model claims it just searched the web but the tool panel is empty, the claim doesn't match what happened. Tool calls aren't the only honest source — base training, keystones, and the rest of the conversation all count — but anything the model says it did, you can verify it actually did.

A close-up of a Memorandai response node's tool-call panel showing three sequential tool calls bundled in a single execution. The panel header reads 'Tools (3)' with a green wrench icon and a Hide toggle on the right. A summary row underneath shows 'Tools: 3' and a green checkmark with the count '3' confirming all three calls succeeded. The Tool Details list below has three rows — keystone-list (collapsed), search_self_memory (expanded), and get-system-info (collapsed) — each with a green check and a chevron caret. The expanded search_self_memory row reveals the exact arguments JSON ({"query": "bundling"}) and a Formatted/Raw result toggle with Formatted active, showing two snippets of returned memory text underneath. Yellow flow-canvas edges enter the node from the left and continue out the right.

A node for everything

Purpose-built components you connect into workflows.

Input Nodes

One per provider. Type a prompt, pick a model, send.

Response Nodes

Streaming markdown output with tool call visibility.

Text Context

Attach reference material to any prompt in the chain.

Image Context

Feed images into multimodal models for visual analysis.

Code Files

Syntax-highlighted source code with version tracking.

Script Nodes

JavaScript execution with pipeline helpers built in.

Shell Output

Run system commands and capture results on canvas.

Comparison

Side-by-side model response analysis with annotations.

Chart

Visualize data with bar, line, area, and pie charts.

Image Generation

DALL-E 3, GPT Image 2, Imagen 4, or Grok Aurora on the canvas.

Custom Nodes

Use the in-app Custom Node builder to create your own nodes.

Node Templates

Load built-in templates or create and save your own workflows.

Script nodes & pipelines

Write JavaScript that chains through your workflow. Transform data, filter results, call APIs, or build full automation sequences without leaving the canvas.

// Pipeline helpers built into every Script node
const data = await upstream();   // read connected input
const result = transform(data);
output(result);                // send downstream
await runDownstream();         // trigger next node

Agentic operation with Claude Code

Users who are comfortable with more advanced AI solutions like agentic workflows can access several different Memorandai MCP servers that give Claude Desktop and Claude Code a different range of tooling, allowing them to take direct actions in Memorandai for you. Agentic decisions and external MCP activity are surfaced in the Agent Hub section of the app, while a specialized selection of nodes are available in the Agent node menu in Flows.

Decision Points

Present choices and wait for human input before proceeding.

Handoffs

Structured task delegation with step-by-step instructions.

Session Summaries

Capture decisions, files changed, and next steps automatically.

Templates

Claude Code can create new custom templates externally for reusable workflow patterns.

Memorandai's Agent Hub view, Activity tab — an audit log of requests tagged with a provenance actor (MCP bridges, external AIs). Five-tab navigation (Decisions, Collaboration, Scratchpad, Questions, Activity active) sits below the page header. Each log entry shows its provenance actor badge (llm-agent in blue for in-app agentic activity, external-ai in orange for external MCP partners like Claude Desktop), HTTP method, endpoint path, response status, and timing. One entry is expanded inline showing its full timestamp, source (external-mcp), and client ID. Filter controls for actor and time range, plus Add External AI / Refresh / Clear log buttons, are visible at the top of the log.

Classic Mode

While certain advanced functionality is only accessible in Flows mode, a "Classic" linear chat interface is available as an alternative to the canvas. Same data, same models, same memory -- just a different shape for the conversation. Toggle between Canvas and Classic with one click.

Memorandai's Classic mode — the linear chat alternative to Flows. The top bar shows the project selector (Test Roundtable with Local Model), a Canvas/Classic mode toggle (Classic active), and a model picker. The conversation thread displays the user's prompt at the top, followed by a Roundtable response with the multi-section structure (Roundtable Synthesis, Brief Overview, Points of Agreement, Points of Disagreement / Divergence) — same data shape as the Flows canvas Roundtable, just laid out vertically as a chat thread. A message input at the bottom shows the next chat turn ready, with Claude as the active model.

Ready to build your first workflow?

Start with a single node. Connect from there.

Get Memorandai