NEXTOP AGENTIC WORKERS STUDIO
project: loading...

Click cards to select them.

      New worker.

      ANY TIME
      1

      IDENTITY

      2

      OUTPUT TERMINALSat least one

      Each terminal starts with a simple text schema; edit the JSON Schema here or in worker.json. The table shows what the program checks (blue: refused when broken) and the descriptions the LLM reads (green, optional).
      EXPECTED INPUTS start worker only
      Used when a block of this worker has no incoming line (a start worker): it gets the run input as its first input section, and these inputs define the flow input (the RUN form fields; required unless unchecked). Elsewhere the lines decide the input.
      3

      TOOLSshared tools from tools/

      Tool notes: how this worker uses each tool (the same tool can be used differently by each worker); sent with tasks and questions. Empty notes on a chosen tool can be saved (the structure check warns); notes on the program's tools (grey) are optional.
      4

      DOMAIN KNOWLEDGEpermanent

      IN THIS WORKER'S KNOWLEDGE

        None yet.

        DOCUMENTS IN THE PROJECT'S docs/ FOLDER
        Tick files or folders to add (a folder takes all .md files in it).
        FILES FROM ELSEWHERE
        .md only, no PDF (convert first). Picked files are copied into the worker's knowledge.
        On save the program builds the map and the index; save finishes only after both.
        5

        LIMITSoptional overrides

        6

        POST-CHECKchecks.py

        Optional checks.py in the worker folder: it checks every saved output and refuses it until its rules hold. Write it by hand or with an AI coding assistant.
        7

        PERSONApersona.md

        Write it last, from everything above.

        MISSIONmission.md

        Sent with assigned tasks only; open questions (ASK) get the persona alone.

        PRE-CHECKan LLM reads this worker's saved parts and lists where they no longer agree

        Reads the saved files (save first). Writes workers/<id>/inspect.md for fixing, e.g. with an AI coding assistant; changes nothing else. Default model and check items: SETTINGS > PRE-CHECK.
        SAVING0%

          Creates workers/<id>/worker.json, persona.md and knowledge/. Fill the details later in the files (vibe coding).

          idle Open in RUNS
          run input
          run input (JSON object, goes to the start functions) back to the input form

          No start block: this flow takes no input. A start function (FUNCTIONS > + New function > start function) or a worker block with no incoming line (a start worker) takes the run input.

          
              

          Lines 100%

          STRUCTURE CHECK

            Click cards to select them.

                New file server.

                Saving gives this file server's tools to everyone on its share lines.

                BASICS

                SHARE LINES

                When a block of the sender ends, the files it owns go to the receivers, who become owners too. The FLOW canvas shows the same list as dashed lines (bottom point -> top point). Functions send with FILES_OUT = True and receive with FILES_IN = True; files the user gives come in through the start function. A share function (file_servers/<name>/share.py, written in code) may pick which files go to whom.

                CHECK

                • Save to run the check.

                TOOLS

                WORKERON THE SHARE LINES ASTOOLS
                View file_server.json
                
                        

                FRAMEWORK TOOLS

                Added by the framework; not files. File server tools go to everyone on its share lines; knowledge tools come with domain knowledge. Grey chips: added by the framework. Green chips: shared tools the worker lists.

                WORKERTOOLS

                A shared tool is a Python file in tools/. This form writes the name, description, parameters and TIMEOUT; you write run() in the file afterwards.

                PARAMETERS
                NAMETYPEREQUIREDDESCRIPTION

                A function is a flow block that runs Python code (functions/<name>.py) without an LLM. It takes any number of incoming lines (one inputs key each) and has output terminals like a worker.

                Click cards to select them.

                  This form writes the name, description, outputs and TIMEOUT of functions/<name>.py; you write run() in the file afterwards.

                  INPUTS (EXPECTED)
                  A note of what the function expects; the lines on the FLOW tab decide what really comes in.
                  OUTPUT TERMINALS
                  Each output starts with a simple text schema; edit the JSON Schema here or in the file later. No outputs = an end function (no output point); a start function needs at least one.

                  Pick a run on the left.

                  ABOUT

                  project.json

                  PROJECT

                  DEFAULTS
                  Empty = framework default (shown in grey). Workers may override some of them.
                  KEYVALUE

                  PRE-CHECK

                  Worker pre-check (PRE-CHECK on a worker, nextop inspect): an LLM reads one worker's parts and lists where they no longer agree. Saved in project.json as inspect.

                  CHECK ITEMS
                  One item per line. Empty: the default list for the project language (shown in grey).

                  KEYS

                  Values live in .env; this page shows only whether they are set. Put key values in .env (never here).

                  KEY NAMEENV FILESTATUS

                  One card per LLM server (adapters.llm): its connection, and the models it serves. Models listed here appear in every model choice (workers, file servers, index) as "model (server)". Empty fields are left out.

                  Embeddings first, then the indexes that use them. Empty fields are left out.

                  Any other adapter kinds in project.json. Empty fields are left out.

                  CHECK

                  • Save to run the check.
                  View project.json