The all-in-one, open-source backend platform purpose-built for AI coding agents — giving your agent database, auth, storage, edge functions, a model gateway, and compute to ship full-stack apps end-to-end.
InsForge is an open-source Backend-as-a-Service designed from the ground up for AI-native development workflows. Unlike traditional BaaS platforms built for human developers clicking through dashboards, InsForge exposes its entire backend through an MCP Server and CLI that coding agents can operate directly — reading schemas, running migrations, deploying edge functions, and configuring auth providers without leaving the agent loop.
The platform bundles every primitive a full-stack app needs: a managed PostgreSQL database with pgvector support for embeddings, S3-compatible object storage, a multi-provider authentication system (including OAuth2 with PKCE), serverless edge functions powered by the Deno runtime, WebSocket-based realtime channels, and an OpenAI-compatible AI model gateway backed by OpenRouter routing across hundreds of models. A dedicated Memory API layered on pgvector lets agents store and semantically recall structured facts, decisions, and preferences across sessions.
InsForge ships as a single Docker Compose stack and can be running locally in minutes. Multiple isolated project instances can run on the same host by using different port configurations. For teams that want to skip infrastructure management entirely, a cloud-hosted version is available at insforge.dev, with one-click deployment templates also available on Railway, Zeabur, and Sealos.
The monorepo is organized as a Turborepo workspace with a Node.js/Express backend, a React/Vite frontend dashboard, a Deno-powered edge function runtime, shared Zod schemas, and an official TypeScript SDK published to npm as @insforge/sdk. The project has released 48 versions since its July 2025 launch and ships new releases at roughly weekly cadence, with over 438 commits per month reflecting rapid iteration.
Architecture InsForge is structured as a layered monolith with clear vertical separation of concerns across domain modules. The backend follows a routes → services → providers → infra pattern: HTTP routes handle request validation using Zod schemas from the shared-schemas package, delegate business logic to singleton service classes, and reach external systems through provider abstractions that isolate third-party dependencies. Cross-cutting concerns — authentication middleware, rate limiting, error handling — are centralized in an Express middleware layer. The Deno edge function runtime runs as a separate process with its own isolated worker model, communicating with the main backend through an internal API, which cleanly separates sandboxed user code from the trusted host environment. A shared Zod schema package enforces a single type contract across the backend, frontend, and SDK without code duplication.
Tech Stack The backend is TypeScript running on Node.js with Express as the HTTP framework, organized as a Turborepo monorepo workspace. PostgreSQL serves as the primary database with pgvector extension for embedding storage and similarity search. Authentication is built on Better Auth with a custom OAuth2 PKCE layer. Object storage uses a local S3-compatible implementation with optional CloudFront CDN integration. Edge functions execute inside isolated Deno workers with AES-GCM secret decryption at runtime. The AI model gateway routes requests through OpenRouter using an OpenAI-compatible API surface. The React/Vite frontend dashboard communicates with the backend over REST. Real-time functionality is handled by Socket.io with a custom channel and permission model layered on top. The npm SDK (@insforge/sdk) is built with tsup and published from the packages workspace.
Code Quality The codebase demonstrates solid engineering practices with extensive test coverage — over 129 unit test files and separate integration test suites, all run through Vitest with V8 coverage reporting. Tests are executed sequentially with single-worker pooling to avoid database conflicts, and a shared setup file bootstraps test fixtures. Service classes use the singleton pattern with explicit initialization methods, and errors are typed through a centralized AppError class with named ERROR_CODES constants. Zod validation is applied at every API boundary using schemas shared from a dedicated package, eliminating a category of runtime type errors entirely. ESLint with TypeScript-eslint and Prettier enforce consistent style. The TypeScript configuration uses Astro’s strict preset, and exhaustive switch checks on discriminated unions (particularly in the AI provider router) prevent silent enum drift.
What Makes It Unique InsForge’s most distinctive technical decision is treating the MCP Server as a first-class interface rather than an afterthought — every backend primitive is designed to be operable by a coding agent, not just a human developer. The Semantic Memory API is genuinely novel for a BaaS: it layers pgvector similarity search with an LLM-assisted deduplication pass, using offline-tuned recall thresholds (0.45 for retrieval, 0.5 for reconciliation) and a typed memory kind system (fact/decision/preference/reference) rather than a generic key-value store. The edge function isolation model — running user-submitted TypeScript inside per-request Deno workers with AES-GCM encrypted secret injection — provides a security boundary that most self-hosted BaaS alternatives leave to the operator to implement themselves.
InsForge is released under the Apache License 2.0, a permissive open-source license that allows commercial use, modification, distribution, and private use without any copyleft requirements. You are free to run InsForge as the backend for a commercial product, modify the source, and distribute modified versions, provided you retain the original license and copyright notices. There are no open-core restrictions or source-available gotchas — the full platform, including the MCP Server, edge function runtime, model gateway, and memory API, is Apache-licensed.
Running InsForge yourself requires Docker and Node.js. The standard deployment is a Docker Compose stack that brings up the Express backend, React dashboard, Deno edge function runtime, and a PostgreSQL instance with pgvector. You are responsible for managing the host machine, keeping Docker images updated, handling database backups, and scaling the stack as traffic grows. Multiple isolated projects can run on the same host by using different port configurations per compose stack. For production deployments, you will want to configure trust proxy settings, set strong JWT secrets and API keys, and wire up external object storage if S3-compatible storage beyond local volumes is needed. The operational burden is moderate — the compose file encapsulates most of the complexity, but ongoing maintenance of the PostgreSQL instance (backups, vacuuming, version upgrades) falls on the operator.
The cloud-hosted version at insforge.dev offloads infrastructure management entirely: no Docker to maintain, automatic backups, managed PostgreSQL with pgvector, and a dashboard for monitoring usage and logs. The trade-off is that the cloud tier removes full data portability and introduces usage-based pricing for compute and storage beyond free tier limits. One-click deployment templates on Railway, Zeabur, and Sealos offer a middle path — managed hosting without the insforge.dev lock-in, though still dependent on those platforms’ pricing and availability guarantees.
No Code Platforms · AI Development · Developer Tools
Visual LLM workflow platform with RAG pipelines, agent capabilities, and model management for building production AI applications.
Developer Tools · Game Development · Design Tools
Free, MIT-licensed 2D and 3D game engine with one-click multi-platform export and no royalties.
Developer Tools · Databases · Search
The open-source Postgres development platform that replaces Firebase with authentication, real-time APIs, edge functions, storage, and vector embeddings — all built on PostgreSQL.