DocsGPT's visual workflow builder automates document processes without code: AI agents, code, shared state and If / Else branches, with human approval where it matters. Runs start from chat, the API, a webhook or a schedule, and each one can hand back the files it produced.
start · ai agent · set state · if / else · code · note · end · shared state
Last updated:
Drag nodes onto the canvas and connect them. Each AI Agent node carries its own model, prompt, tools and knowledge, can work as a Classic or a Research agent, and can return structured JSON for the next step to use.
Notes sit on the canvas beside the steps they explain, and undo and redo keep editing safe.
start · ai agent · set state · if / else · code · note · end
Start begins every run, from chat, the API, a webhook or a schedule. AI Agent runs an LLM step with its own model, prompt, tools and knowledge. Set State writes shared variables. If / Else branches the run, in Simple mode or with CEL. Code runs a Python script in the sandbox. Note explains the flow on the canvas. End returns the result.
| Does | Set up | |
|---|---|---|
| Start | where every run begins | runs from chat, the API, a webhook or a schedule |
| AI Agent | an LLM step | its own model, prompt, tools and knowledge; Classic or Research; structured JSON output |
| Set State | writes shared variables | values computed with CEL |
| If / Else | branches the run | Simple mode, no code, or CEL expressions; always an Else |
| Code | runs a Python script in the sandbox | reads the state; files it writes become artifacts |
| Note | explains the flow on the canvas | for the people who maintain it |
| End | returns the result | the answer the run hands back |
If / Else nodes branch in Simple mode: pick a variable, an operator and a value, and name each branch. Every node has an Else, so no run falls through.
Switch to Advanced for CEL expressions such as confidence < 0.8 && amount > 10000. CEL is side-effect-free, so given the same state a route always goes the same way: a fact you can read, not a model's judgement. The shared state is a set of variables every node can read and write.
simple · advanced (cel) · named branches · else · shared state
Approval gates hold a response before it is delivered or a tool call before it executes. The reviewer sees the pending item and approves or rejects it in one click; every decision is logged.
The same gates apply across the platform. See guardrails and human-in-the-loop.
gate: response | tool execution · approve / reject · logged
Preview runs a workflow beside the canvas and lights up each step as it executes, with the execution details and any files the run produced, such as a spreadsheet or a memo.
Every run is kept in the agent's Logs with its trace, so you can see which step took which branch, alongside message, token, tool and scheduled-run statistics. Workflows export and import as files, so a flow can move between deployments or into version control.
Workflow runs, shared state and run history stay in your own database. Code nodes run in the same sandbox as artifacts and code execution, inside your network.
Four deployment postures, one code base. Compare deployment options →
Inside your network
managed · on-premises · air-gapped
Cloud
Arc53 operates the engine and sandbox; runs and history are stored in Arc53's cloud.
Air-gapped
Same engine, state and history in your database; AI Agent nodes use a local LLM; Code nodes use the self-hosted sandbox. No external network.
24 h → 16 min
A large Central European commercial bank processes 10,000+ inbound documents a day; time-to-action fell from 24 hours to 16 minutes, with over 99.5 % classification accuracy, running fully air-gapped on the bank's existing NVIDIA T4 GPUs.
Source: EU bank, name under NDA
1–2 days → ~5 min
Transfer-document review that took 1–2 working days (P95) now completes in about 5 minutes, with 12 deterministic and agent checks and human approval for uncertain cases.
Source: Bank (POC), name under NDA
Documentation
Next: agent types and tools · guardrails and approval gates · generated files and the sandbox · banking and financial services · deployment options · pricing
A 30-minute demo on your use case, or a free on-premises review with an engineer.