InsForge: Open-Source Backend Platform for AI Coding Agents

InsForge (YC P26) is an open-source backend platform designed specifically for AI coding agents. Instead of agents juggling dashboards and manual config, InsForge gives them a CLI and skills to handle all backend/infra tasks end-to-end. The project is available on GitHub under Apache 2.0.
How It Works
Coding agents interact with InsForge through two interfaces:
- MCP Server (self-hosted and cloud): exposes InsForge operations as tools for any MCP-compatible agent.
- CLI + Skills (cloud only): a command line interface paired with Skills that agents invoke directly from the terminal.
Both allow agents to read backend context (docs, schemas, logs) and configure primitives (deploy edge functions, run migrations, create buckets, set up auth providers).
Core Products
- Authentication – user management, auth, sessions
- Database – Postgres relational database
- Storage – S3 compatible file storage
- Model Gateway – OpenAI compatible API across multiple LLM providers
- Edge Functions – serverless code on the edge
- Compute (private preview) – long-running container services
- Site Deployment – build and deploy sites
Agent-Specific Features
- Backend Branching: branch entire backend (DB, auth, storage, functions, schedules). Agents work on a branch, you review diffs, merge or discard. Inspired by Neon.
- Server Telemetry: agents read logs, CPU, memory, disk to find spikes and root causes.
- Debug Agent: every project gets a dedicated debug agent. Your coding agent asks questions like “why deployment fail?”, the debug agent runs diagnoses and proposes fixes.
- Backend Advisor: scans your backend daily for security and performance issues, proposes remediations, sends to your coding agent.
Quickstart (Self-Hosted)
git clone https://github.com/InsForge/InsForge.git
cd insforge
cp .env.example .env
docker compose -f docker-compose.prod.yml upThen open http://localhost:7130 to connect the MCP server. After setup, ask your agent: "I'm using InsForge as my backend platform."
Cloud version is available at insforge.dev.
The team notes they started InsForge because manual MCP setups had issues: tools pre-loaded into context, large payloads (>10k tokens), and missing capabilities like telemetry and configs. By putting everything into a CLI and teaching agents via skills, they avoided those problems.
📖 Read the full source: HN AI Agents
👀 See Also

Caliby: Open-Source Embedded Vector Database for AI Agents with Hybrid Text+Vector Storage
Caliby is a C++ embedded vector database with Python bindings (pip install caliby) that supports HNSW, DiskANN, and IVF+PQ indexes, claims 4x performance over pgvector, and natively stores text alongside vectors for AI Agent/RAG use cases.

Agnost AI Launches: Product Analytics for Chat and Voice Agents
Agnost AI reads production conversations to surface behavioral failures like rageprompting, repeated rephrasing, and hidden feature requests.

Benchmark Shows CLI Tool Reduces Claude Code Token Costs by 32% Through Structural Navigation
A developer built a Rust CLI tool that gives Claude Code agents structural navigation commands like 'show me a 180-token summary of this 6,000-token class.' Benchmarking on Sonnet 4.6 across 54 automated runs showed 32% lower cost per task and 67% more code edits per session.

OpenClaw Skill Reduces Accessibility Tree Tokens from 600K to 1.3K
A developer built an OpenClaw skill that uses ML-based element ranking to prune accessibility trees, cutting slickdeals.com from ~598K tokens to ~1.3K tokens by keeping only the top ~50 actionable elements.