- Add `advanced` to `SchemaField` and pass it to `json_extra`
- Add `advanced` to `BlockIOSubSchemaMeta` type
- Update `CustomNode`, so that:
- non-required advanced inputs are hidden
- non-advanced and required inputs are always shown
- Add minimize/maximize button in the corner of modal to make it significantly larger and centered
- Add copy button to copy all text
- Add optional `title` to display as a modal header
- fix type propagation by `AppService.run_and_wait(..)`
- fix type propagation by `@expose` and add note
- fix type propagation by `wait(..)` in `.executor.manager.execute_node(..)`
- fix type propagation by `wait(..)` in `.executor.manager._enqueue_next_nodes(..)`
- remove unnecessary null checks for `.data.graph.get_node(..)`
- fix type issue in `ExecutionScheduler`
- reduce use of `# type: ignore` in `.data.execution`
- reduce usage of `# type: ignore` in `.executor.manager`
- reduce usage of `# type: ignore` in `.server`
- reduce usage of `# type: ignore` in cli.py
- update `pyright` to v1.1.378
* standalone websocket server
* add websocket url
* wip: talk to ws directly
* rename to api server
* dockerfile and queue
* fix paths
* update poetry lock
* helm charts for websockets
* create seperate deployments for websockets and rest server with redis queue for async comms
* delete duplicate queue
* add depends in ws_api
* singleton for conn manager
* update from review
* fix CI
* address feedback
* update readme
* update docker file and add migration step in readm
* ad watch
* add step to copy example env file
* put connect back in
- Update styling and use tailwind more
- Add `react-toast` dependency
- Fix output button not changing checked state on execution
- Make status a badge in node's corner
- Rename `output_data` to `executionResults` and store multiple results with execution UUIDs
- Add `DataTable` component that displays execution outputs
- Outputs can be copied and there's a toast displayed as a feedback
Issue 1:
Input text field cursor keeps moving to the end of the text.
Try to type "Hello World!" into the input text. Then try to type "some string" in the middle of the "Hello" and "World".
Issue 2:
History should only tracks on the input box onBlur/onLeave
Try to type a "longcharacters" and try to undo it, the undo is removing 1 character at a time, polluting the history, and make the undo pretty much unusable.
Issue 3:
KeyValue & ArrayInput is non-undoable.
Try to add key-value or add an entry to the list, it doesn't undo the value, but you need to click as many number of entries being added to make the undo work again
* Feat(Builder): Add first guide tutorial
* added more steps + some fixes
* added local storage to fix starting every time going to build
* update copy & paste to support mac
* small fix
* Prettier fixes
* Added "Skip Tutorial" button to first step
* some fixes based on requests
* revert camelCase change
* add ability to use url to reset tutorial
* prettier
* Added Tutorial button next to tally
* prettier
* change pinBlocksPopover to setPinBlocksPopover
* fixes + update + prettier
* made the resetTutorial url dynamic
* force to /build on reset tutorial
* fix renaming
* prettier
Update ReactFlow to version 12 and split `Flow.tsx` into `useAgentGraph` hook that takes care of agent state and API calls to the server.
- Update ReactFlow to v12 ([migration guide](https://reactflow.dev/learn/troubleshooting/migrate-to-v12))
- Move `setIsAnyModalOpen` to `FlowContext`
- Make `setHardcodedValues` and `setErrors` functions of `CustomNode` and utilize new `updateNodeData` ReactFlow API
- Fix type errors
- `useAgentGraph` hook
- Take care of all API calls, websocket, agent state and logic
- Make saving and execution async and thus more consistent and reliable
- Save&run requests are state
- Wait for node ids to sync with backend reactively
- Queue execution updates
- Memoize functions using `useCallback`
### Background
Currently, there is no way to construct the output of nodes into a composite data structure (list/dict/object) using the builder UI.
The backend already supports this feature by connecting the output pin to the input pin using these format:
* <pin_name>_$_<list_index> for constructing list
* <pin_name>_#_<dict_key> for constructing dict
* <pin_name>_@_<field_name> for constructing object
The scope of this PR is implementing the UX for this in the builder UI.
### Changes 🏗️
<img width="765" alt="image" src="https://github.com/user-attachments/assets/8fc319a4-1350-410f-98cf-24f2aa2bc34b">
This allows you to add more pins in a key value & list input: `_$_` list constructor & `_#_` dict constructor.
### Background
Boolean without default value is a UX problem. It's currently displayed as a toggle and it has no way to describe the `null` value.
So we need to prevent blocks from introducing a nullable boolean.
### Changes 🏗️
Add explicit check to prevent nullable boolean. Fix existing block field that has nullable boolean.
* talking head
* linting
* remove clip id, not needed
* add more descriptive name
* add min requirement to polling attempts and intervals
* add docs and link to docs
* remove extra space
* force new tab
* fix linting
* add did key to .env.template
### Background
We don't have an ordering guarantee on the node execution.
Let's say we have a node that has to execute different data A, B, and C.
The current implementation limits the execution to 1 execution at a time, but there is no guarantee that A, B, and C will be executed in order.
The initial implementation did not have any restrictions, so it used to be A, B, and C executed in parallel
In the current implementation with the per-node constraint, it's A, B, C are executed serially but with no guarantee of ordering.
The scope of this PR is to guarantee that order.
### Changes 🏗️
Guaranteeing the execution per node ordering by avoiding any re-enqueue mechanism. If there are two executions run in the same node, the first one will be executed and the other will block. The blocking mechanism is indeed sub-optimal, the performance improvement can be done later (a follow-up issue will be added).