In Review
18Under consideration
Integration to ZeroEntropy
Integrate with ZeroEntropy API for Vectorization, Ingestion and Query
Add integration to Mem0
Support memories additional and retrieval from managed Mem0 platform
MCP-Rollback Tools
Rollback Tool Suite for Dev MCP
Support Subflows / Stages within a flow
Customer want to build multiple agentic flows with different payload and stages within the same flow There are various options we can provide this - Suspend - Resume logic Dedicated Followup Node which sends back the response and takes the input then resumes the flow where it left
Option to Enable or Disable Auto-Format in IDE
Feature - When nodes are added or removed in the flow UI, the layout automatically re-formats. Why - This causes unexpected changes in the flow's appearance, leading to a frustrating user experience. Where would it help most - This feature gives users control over the flow formatting behavior, improving usability and reducing frustration. Extras - Introduce a toggle switch that allows users to choose whether they want auto-formatting enabled or disabled. Auto-format ON: The flow automatically adjusts layout when nodes change. Auto-format OFF: The flow layout remains unchanged, preserving the user’s manual arrangement even when nodes are added or removed. Sample format Image before adding END Node Format image after adding END Node
Add Evals Functionality
Types of Evals - On Test Debug On Deployments Compare On Set of Logs Realtime Sampling Options Build Inhouse with AI Integrations Connect with Langfuse and do it cross directional
Ability for a support team member to join the chat with a user interacting with the support agent
It would be nice if a member of the Customer Support team could jump in to a live chat when a user interacts with the support widget, and isn’t able to get their question fully answered.
Ability to export the code and deploy on other platform ( BYOC )
Support for code export like other platform like webflow
Fetch user's data using project ids
Feature - Get the user’s data using project ids Why - This will help in identifying the users that are affected during the potential outage
Bulk Retry for Failed Flows
Problem: After an incident, multiple flows may fail due to the same underlying issue. Currently, we have to retry each failed flow individually from Studio, which is time-consuming and inefficient when there are a large number of failures. Request: Add a Bulk Retry functionality in Studio that allows users to select multiple failed flow executions and retry them together in a single action. Expected Benefit: This would significantly reduce manual effort and make incident recovery faster, especially when a large number of flows are affected by the same incident.
Planned
36Committed and queued
Support for Local Development
An open-source SDK to address developer adoption barriers and vendor lock-in concerns that are currently limiting growth. Problem Developers resist Lamatic's no-code/low-code platform due to limited library support, lack of local development capabilities, missing CLI tools, and weak version control integration Enterprise and startup customers worry about vendor lock-in and business continuity if Lamatic changes pricing or discontinues service Competitors offering open-source alternatives have an advantage Proposed Solution Build a standalone, open-source SDK enabling developers to: Develop AI agents/systems locally in their preferred IDE Test and debug offline without cloud dependency Use Git for version control across all project components Create custom integrations and capabilities Deploy anywhere (on-premise, private/public cloud, or Lamatic-managed) Compile flows into executable code via CI/CD pipelines Core Architecture YAML-based configuration system with: Pre-built node libraries for AI components Containerized runtime (Pod + Core) executing configurations Compiler for optimizing YAML into executable workers Local web editor accessible via localhost CLI for all development operations Multiple invocation methods (API, webhooks, widgets)
Full Refresh option in Force Sync
Users currently struggle and have to perform multiple steps when indexing a new schema in the vectorDB. Current Steps - Update the metadata schema Change the mode the full refresh deploy the flow Schedule Runs Data indexed Proposed Steps - Option in Jobs > Force Sync > Full Refresh Automatically updates flows, Deploys it, Trigger a Job Return back the flow to Incremental
Enhance Flow Comparison
Multi-tab Support and Flow Comparison Enable right-click to open flows in new tabs for easy comparison Support multiple tabs within the main window (similar to how Linear handles native apps) Display full flow names - currently get cut off/truncated in the UI, making it difficult to distinguish between flows Allow side-by-side visual comparison of different flow versions Consider implementing an Electron wrapper for native app experience with better tab support Cross-Environment Flow Navigation Ability to change environments within a single flow view - users want to see what the same flow looks like in staging while viewing production Maintain flow state when switching environments instead of returning to flow home screen Use flow slug associations to link flows across environments (dev/staging/production) Reduce manual effort required to promote workflows from dev → staging → production
Spaces for static file uploads and usage
Ability to upload files and use them in flows (to, among other things, eliminate the need to find a place to host files either for testing or production purposes). For example, PDF upload for training or CSV for instructions.
Support For dynamic File References in Flow Config
The nodes are embedded inline with formatting artifacts (hello\\nhello), apparent test content, and no versioning metadata. Editing it requires understanding the entire config file, and changes produce noisy diffs that obscure actual prompt improvements. Before: prompts: - id: 187c2f4b-c23d-4545-abef-73dc897d6b7b role: assistant content: >- hello hello Important:please provide the response in wrap text format... After: prompts: - id: rag_system_prompt_v1 role: assistant # Edit prompts in /prompts/rag_system.md — supports preview & versioning contentRef: ./prompts/rag_system.md version: "1.2.0" lastUpdated: "2025-01-15" Developer Impact: Prompts become first-class artifacts — reviewable in PRs, A/B testable, and editable without touching pipeline config.
Break Node and End Node
Problem: No clear way to terminate workflows early or break out of loops based on conditions, forcing users to create workarounds with maximum iteration counts. Use Case: When API calls fail or specific conditions are met, users want to cleanly exit workflows or break out of polling loops without continuing unnecessary iterations. Proposed Solution: Implement dedicated break and end nodes that allow conditional workflow termination and loop breaking.
Automated Response Evaluation System (Graders)
Problem: Manual grading of responses does not scale and is inconsistent. Use case: Systematically assess response quality across large test suites. Functionality: Grader agents with configurable rubrics, batch “grade all” execution, and integration with logs for historical evaluation.
Connect with External Vector DB
Support Vector DB Weaviate PG_Vector PineCone Additional MongoDB ChromaDB Qdrant
Vector Search as Tool
Support Inbuilt vector search of vectorDB as tool call for AI Node
A/B Testing Analytics and Reporting in Logs
Problem: A/B experiments lack clear labeling and result visualization. Use case: Make data-driven decisions about prompts, flows, and model choices. Functionality: Experiment labels on runs, statistical analysis with confidence intervals, dashboards for variant comparison, and filtering/grouping by experiment tags.
In Progress
6Actively being built
Support Custom Endpoint when Adding Models
Add optional support to add custom API base URL when adding new model credentials so that customer can choose what region / instance they want to connect to
Token Streaming Support in SDK
Problem: Users perceive latency because initial tokens are delayed visually even when backend completion is fast. Use case: Improve UX for long-running tasks in web apps by showing progress immediately. Functionality: Dual token streams (output and process), a "processing response" JSON key for intermediate node updates, and an async streaming API that can pipe data directly into React components.
Support variables in zod schema of generate json node
There should be option to add zod schema as a variable in generate json node
Extend Request Metadata to support session info
Extend request metadata with optional fields for session enrichment: userId, sessionId, and orgId Location and client context: IP address, coordinates, country, city, device, and browser Automatically capture available fields for requests made through widgets and client-side SDKs. Allow developers to supply or override supported fields when needed. Make each field filterable in logs and reports so teams can investigate requests by user, session, organization, or client context. Acceptance criteria Requests can store the optional user, session, organization, and location/client fields. Widgets and client-side SDKs populate available fields automatically. Logs and reports support filtering by each field. Missing fields do not block requests.
Show Testing Meta Data should show some more details
Show Testing Meta Data should show some more details • show timings and latency • Parallel execution doesn't show accurate timings instead mimic the timings from a sequential execution
Make Logs OTEL Compatible
So that we can connect and export it to other log systems
Completed
122Recently shipped
Model Preference & Fallback
A dedicated UI to choose model preferences on project level. There will be different categories of models ( Tabs ) for Text | Embedding| Chat | Image | Audio | Transcript For Each category customer can create 5 Settings and reorder them. In each settings can choose 4 tabs ( Basic | Fallback | Best | A/B Test ) When using model selection in a workflow node, choose a preferred setting options which will populate the UI with credentials and model selection. ( By Default the first preference will be chosen or create a new preferences
Support Speech via widgets
Use web API to give real time TTS and STT on search and chat widget
Dispay Test Run Stats in Debugger
Problem: Developers must manually compute speed and token metrics from logs. Use case: Quickly benchmark and optimize workflows against performance targets. Functionality: Real-time display of execution time, token usage, accuracy indicators, TTFT, throughput, and benchmark comparisons within the debugger.
Add support for S3 action Nodes
Currently we only support S3 as a trigger. Add support to use it as a action node as well.
Native integration to mySQL
Requirement from Pat at co-dev to connect to mySQL (on PlanetScale).
Add MCP Support
Add Support for connecting custom MCP Server in the Tools section of each AI node and integration Section Desired Flow Adding Directly Go to Integration Choose Add Integration Choose MCP Add MCP Server Credentials Via AI Node Choose Any AI Node Go to tools > Add Tools Choose MCP Select MCP Server / Setup New Via Node Create a new MCP Action Node Choose MCP Node Choose Credentials Define Action to perform Find details here - https://www.perplexity.ai/search/how-can-lamatic-ai-build-mcp-i-9.Gp.fxuQ_WWHzBBqlqFUA
Add tries for nodes executions
Add a global level property that sets how many times a given node can retry. Use a default value set on project setting level then it could be over-ridden for every node seperately.
Show Only Compatible Test Cases for Current Flow
Automatically filter test cases and only show the ones that match the current flow’s input schema. @cwhiteman
Webflow Integration
Add Webflow as Data Loader to Sync content into Lamatic
Multi-Branch Execution for Parallel Processing
Problem: Classifier and condition nodes can only route to one branch at a time. Use case: Execute multiple analyses in parallel (e.g., email, image, social checks) based on classification. Functionality: Multi-branch execution support, per-branch on/off toggles, and optimization for concurrent rather than sequential processing.