GitHub - dksingh1997/Toolflow: Manage tool responses in AI agents. Works with Composio, Pipedream, Custom APIs

2 min read Original article ↗

AI-Native Tool Calling Framework — Build intelligent agents that interact with tools via natural language, with minimal context bloat and full developer control.

⚠️ The Problem

Modern AI agents rely heavily on tool calling to fetch data or perform actions. But in practice:

  • 🔥 Tool responses are too large, flooding the LLM context window.
  • 🤯 Developers have no control over what part of the API response is used.
  • 🧩 Mapping tools to agents is hard across different APIs and use cases.
  • 🐘 Most frameworks assume monoliths — no modularity or control for devs building production systems.

"Fetching Gmail emails returns hundreds of fields. I just need subject, from, and snippet!"
– Every AI dev, ever

✅ The Solution: Toolflow

Toolflow is an AI-native middleware layer to connect APIs and tools to LLM agents, with:

  • 🔍 Minimal context payloads: Filter what actually goes to the LLM.
  • 🛠️ Composable tool registry: Define tools + actions as modular objects.
  • ⚙️ Custom modes:
    • Light – Send only minimal fields
    • Full – Send entire response
    • AI Filter – Auto-trim fields based on relevance
    • Custom – Dev chooses fields at runtime
  • 📦 Plug-n-play TypeScript packages for runtime, registry, and types.
  • ⚡ Works with existing frameworks (Langchain, Flowise, AutoGen, etc.)

🧪 Example Use Case

Tool: Gmail.FindEmails

Filtered response :

{ "data": { "messages": [ { "from": "Dheeraj Kumar", "subject": "Repo check", "snippet": "Hi Dheeraj, Does this repo makes sense ?", "date": "Tue, 17 Jun 2025 12:01:00 +0200" } ], "nextPageToken": "16428802254119354251", "resultSizeEstimate": 201 }, "error": null, "successfull": true, "successful": true, "logId": "log_3XvoX1kpxPM6" }