Skip to content

Langflow vs flowMAS feature parity

This document maps Langflow (v1.10.x) capabilities to flowMAS (web studio + gMAS runtime). Use it to track what we already match and what to build next.

Legend

Status Meaning
✅ Supported with comparable UX
⚠️ Partial / different model / workaround
🔜 Planned or stub only
❌ Not in scope or not applicable to gMAS

Build & canvas

Langflow flowMAS Status Notes
Visual flow editor Workflow canvas ✅ Agents, Start/Finish, edges
Component library (Core + Bundles) Agent + Tool registry ⚠️ Langflow has LLM, vector, parser, etc. components
Flow templates Graph templates API ✅
Import Langflow JSON POST /api/graphs/import ⚠️ Structure preserved; semantics differ
Rename / describe nodes Agent inspector ✅
Group components — ❌ Multi-select group not implemented
Freeze component (skip re-run) — ❌ gMAS re-executes agents each run
Run single component — ❌ Whole graph run only
Inspect component output Inspector + Console ✅ Per-agent I/O and tool calls
Typed ports / color coding Agent edges only ⚠️ No Message/Tool port types on canvas
Dynamic ports — ❌
Component code editor Agent Python in registry ⚠️ Agents are code-first, not in-canvas
Component version updates — ❌
Keyboard shortcuts Partial hotkeys ⚠️
Start / Finish nodes Start / Finish ✅ Replaces Langflow Chat Input / Output

Control flow & routing

Langflow flowMAS Status Notes
If-Else / conditional routing Edge conditions ✅ success / fail / keyword / loop
Keyword branches (If-Else style) Conditional + keyword edges ✅ Cyan edges; Run Settings hints
Loop (list iteration component) — ❌ Langflow Loop over lists; use graph loops
Review / cycle loops Loop edges (↺) ✅ Max Loop Iterations in Run Settings
Parallel branches Run Settings → Parallel mode ✅
Multi-input fan-in (AND join) Multiple incoming edges ✅ Skipped branches unblock join (runtime fix)
Router / dynamic path Adaptive scheduler + edges ✅ Requires Adaptive Scheduler
Timer / delay inside flow — ❌ Not edge delay — use Schedules
Notify / Listen (delayed events) — ❌ Langflow still evolving

Execution & triggers

Langflow flowMAS Status Notes
Run once (Playground / API) Run button + POST /api/execution/run ✅
Token streaming WebSocket events ✅
Run history Runs page ✅
Cancel run API + UI stop ✅
Follow-up / continue run Follow-up API ✅
Background Agents (cron / interval / date) Schedules /schedules ✅ pause, resume, run now; APScheduler
Workflow API v2 background: true Schedules + async runs ⚠️ No generic job queue; scheduled + WS
Webhook trigger (POST /webhook) — 🔜 Event-driven external trigger
Event triggers (Background Agents) — 🔜
Tweaks at runtime Run Settings + per-run config ⚠️ No per-component tweak map
Session ID / chat memory Task memory in Run Settings ⚠️ Different model than Langflow sessions
Global variables in request — 🔜 Langflow v2 globals in JSON body
Sync workflow with timeout — ⚠️ gMAS runs async; poll Runs / WS
Stop workflow by job_id Cancel by run_id ✅

Schedules API (Background Agents analogue)

Action Endpoint
List GET /api/schedules
Create POST /api/schedules
Update PUT /api/schedules/{id}
Delete DELETE /api/schedules/{id}
Run now POST /api/schedules/{id}/run
Pause / Resume POST .../pause / .../resume

Schedule types: cron (5-field), interval (seconds ≥ 30), once (ISO datetime).

Run metadata: scheduled runs set trigger_source: "schedule", schedule_id, schedule_name on persisted runs.

Agents, tools & MCP

Langflow flowMAS Status Notes
Agent component Agent nodes ✅
Tool Mode (component as tool) Tool registry + agent tools ⚠️
MCP server (expose flows) — ❌
MCP client (external servers) — ❌
Composio / bundle integrations Custom tools ⚠️ web_search etc. via tool config

Observability & debugging

Langflow flowMAS Status Notes
Playground (test flow) Run + Console ✅
Component logs Console + Inspector ✅ Click log row for details
Run metrics Runs detail ✅ Tokens, timing
Export flow / run JSON export ✅
Validate before run Validate button ✅ Inline graph validation

API & deployment

Langflow flowMAS Status Notes
/v1/run/{flow_id} /api/execution/run ✅
Flow CRUD API /api/graphs ✅
Agent CRUD API /api/agents ✅
API key auth on webhooks — 🔜
Docker deploy docker-compose.yml ✅
Multi-replica scheduler Single API process ⚠️ APScheduler in-process; no distributed lock

Settings & LLM

Langflow flowMAS Status Notes
Multiple LLM providers Settings → LLM Providers ✅
Per-flow provider override Run Settings ✅
Early stop rules Run Settings ✅ keyword / metadata / custom
Adaptive scheduler / pruning Run Settings ✅ Required for conditional edges
Parallel execution limits Run Settings ✅

Priority backlog (Langflow gaps)

  1. Webhook triggers — POST /api/webhooks/{graph_id} to start a run from external systems (Langflow Webhook component).
  2. Background job queue — optional background: true on run API with job polling (Langflow Workflow API v2).
  3. Global run variables — typed globals on execution request.
  4. Distributed schedules — Redis/DB-backed scheduler for multi-replica API.
  5. Component groups / freeze — lower priority; different execution model.

References