Workflows & Automation22.4ms Reflex$0 / MIT License
9,400 active downloads
ThoughtDAG
Non-autoregressive DAG decision orchestrator for multi-step agent plans
jevproxy // thoughtdag (reflex kernel)
$npx jevproxy run thoughtdag
JevProxy Intercept Resolved(Mode: DIRECT_REFLEX)
22.4ms|Cost: $0.0001|0 reasoning tokens burned
STDOUT • Tool Call Output:Status: 200 OK
agent.dispatch(thoughtdag_schedule)
payload: { "dag_spec": { "type": "object", "description": "JSON graph of nodes and dependency edges" },...
✔ Decision short-circuited in 22.4ms without roundtrip to frontier LLM.
Upstream token bill saved: $0.0240 on this turn.
ThoughtDAG
Compatible with Cursor, Claude Code, Windsurf, OpenCode
TRADITIONAL LLM CALL:UNOPTIMIZED
• Median Latency: 2150 ms
• Cost per Turn: $0.0240
• Mode: Full KV Cache Reload & TTFT Prefill
JEVPROXY REFLEX KERNEL:77x FASTER
• Median Latency: 22.4 ms
• Cost per Turn: $0.0001
• Accuracy: 97.8% deterministic
1-Click CLI Execution
Run this tool accelerated through the JevProxy gateway without manual wiring:
npx jevproxy run thoughtdagOpenAI / Anthropic Tool Schema
JSON SpecificationPaste this schema into your agent tools definition or Cursor extensions:
{
"type": "function",
"function": {
"name": "thoughtdag_schedule",
"description": "Parse execution DAG and dispatch ready nodes to worker agents",
"parameters": {
"type": "object",
"properties": {
"dag_spec": {
"type": "object",
"description": "JSON graph of nodes and dependency edges"
},
"max_concurrency": {
"type": "number",
"description": "Max parallel subagent workers"
}
},
"required": [
"dag_spec"
]
}
}
}Technical Architecture & Usage
DAG Orchestration Without LLM Bloat
When managing multi-agent swarms, deciding which sub-task is ready to execute is a deterministic graph evaluation problem. ThoughtDAG handles graph state transitions in 22.4ms.
Accelerate ThoughtDAG with JevProxy
Get 5,000,000 free decision tokens. Eliminate the 3-second tool freeze in Cursor and Claude Code in under 60 seconds.