Build a coding agent

From a bare LLM loop to a production coding agent — seven repos in the order you'll actually need them, with the why at every step.

7 steps · every link human-reviewed · how we curate →

1. 1
   LangGraph★ 37kagents · orchestration

   Build stateful, multi-actor LLM apps as graphs — durable execution, human-in-the-loop, streaming.

   Why this stepStart with the substrate: a stateful graph with durable execution, streaming and human-in-the-loop. Every hand-rolled while-loop agent eventually reinvents this, badly.
2. 2
   deepagents★ 26kagents · orchestration

   LangChain's batteries-included agent harness on LangGraph — planning, sub-agents with isolated context, filesystem, shell, skills, human-in-the-loop and persistent memory out of the box.

   Why this stepDon't wire planning, sub-agents and file tools yourself — this is the batteries-included harness on top of LangGraph. You write the task, not the plumbing.
3. 3
   CubeSandbox★ 10kagents · local

   Hardware-isolated microVM sandboxes for AI agents — sub-60ms boot, <5MB overhead, E2B-compatible API, self-hosted on your own KVM nodes.

   Why this stepYour agent now runs shell commands and arbitrary code. Give it a hardware-isolated microVM instead of your laptop — sub-60ms boot, E2B-compatible API, self-hosted.
4. 4
   tokensave★ 440coding · local

   Code-intelligence MCP server for coding agents — a pre-indexed semantic graph (libSQL + FTS5) they query instead of grepping: symbols, callers, impact radius in one call. 100% local, 50+ languages.

   Why this stepStop burning context on grep. A pre-indexed semantic code graph over MCP — symbols, callers, blast radius — so the agent asks questions instead of reading the repo.
5. 5
   mirage★ 3.3kagents · coding

   Unified virtual filesystem for AI agents — mounts S3, Slack, Gmail, Postgres and ~50 backends as one tree so any bash-speaking LLM can grep and pipe across services. Snapshotable, embeddable.

   Why this stepReal tasks touch S3, Slack, Postgres, GitHub. Mount them all as one filesystem and your bash-speaking agent needs zero new tool vocabulary.
6. 6
   skillkit★ 1.4kcoding · skills

   Package manager for AI agent skills — install from 400K+ skills across 31 sources, auto-translate between 46 agents' incompatible formats, security-scan on install, sync to every agent at once.

   Why this stepTeach it tricks without prompt-stuffing: install security-scanned skills from 400K+ across 31 sources, auto-translated to your agent's format.
7. 7
   LangSmithevals

   Trace, test and monitor LLM apps in production.

   Why this stepYou can't improve what you can't see. Traces on every run, eval suites before every prompt change — close the loop before you ship.