ACI.dev

Aipotheosis Labs founders Will Lee and Hanyi Du engineered ACI.dev to eliminate the manual glue code required to connect autonomous AI agents with external cloud APIs and developer services. Instead of registering isolated OAuth applications and hardcoding custom wrappers for each vendor, the platform coordinates multi-tenant credential storage, permission controls, and function calling behind a unified gateway. That design gives engineering teams an open-source control plane to run hundreds of integrations across agentic IDEs and autonomous assistants without exposing secrets to external proxies.

Why ACI.dev instead of Rube

Rube operated as a hosted Model Context Protocol hub backed by Composio, routing agent requests, function schemas, and third-party credentials through managed cloud infrastructure until its shutdown. ACI.dev approaches the integration layer as self-hosted software that platform operators deploy directly within their own cloud or private data centers. By managing OAuth tokens, API secrets, and invocation logs on private infrastructure, operations teams avoid external data leakage and maintain compliance boundaries that third-party MCP intermediaries compromise.

The operational model also differs from general-purpose workflow platforms like Pipedream. Pipedream relies on predefined event triggers and visual execution graphs built for human developers, which creates operational friction when autonomous models need to dynamically choose and execute arbitrary actions. ACI.dev avoids static workflow wiring by using runtime meta-functions for search and execution. Rather than injecting hundreds of tool definitions into model system prompts and exhausting token limits, agents discover relevant schemas dynamically and execute function payloads on demand.

What it does well

• Single unified Model Context Protocol container that exposes over 600 third-party developer and SaaS tools to agentic environments.
• Intent-based dynamic tool discovery with dedicated search and execute meta-functions, keeping model prompt context clean and token consumption low.
• Multi-tenant credential isolation with automated OAuth token refreshes and secure secret handling for end-user accounts.
• Natural-language permission boundaries that allow operators to restrict destructive API mutations without writing bespoke middleware filters.
• Detailed invocation telemetry that records function arguments, latency, and response payloads for debugging and security auditing.