# DeepAgents Code Hosting on a Real microVM > Run LangChain's DeepAgents Code (dcode) — the model-agnostic terminal coding agent — on its own always-on Firecracker microVM. Point it at any provider:model via OpenRouter (GLM-5.2, DeepSeek, Kimi), no frontier-lab subscription. 124ms boot, 30-day money-back. - Published: 2026-06-30 · Updated: 2026-06-30 · jurniti - Canonical: https://www.jurniti.com/blog/deepagents-code-hosting LangChain shipped DeepAgents Code with one decision that sets it apart: it doesn't pick your model for you. Point `dcode` at any `provider:model` string and go — GLM-5.2, DeepSeek, or Kimi through OpenRouter for cents, no $20/mo frontier subscription gating the door. The catch is the same one every terminal agent has: it runs real `bash` commands its planner decided on, and on your laptop that's your whole machine on the line. Here's the move: run [DeepAgents Code](/deepagents) on its own always-on Firecracker microVM. Bring whatever model you want, keep a persistent workspace, and let the agent run inside a box it can't escape — booting in about 124ms. ## What DeepAgents Code is DeepAgents Code (the `dcode` binary, shipped as `deepagents-code`) is LangChain's terminal-first coding agent, built on their deep-agents architecture. The headline isn't a model — it's the freedom to bring any model: - **Truly model-agnostic.** A `provider:model` string is the whole config: `dcode --model openrouter:z-ai/glm-5.2`. OpenRouter, Anthropic, OpenAI, Fireworks, Baseten, or local. - **Open weights, no subscription.** Run GLM-5.2, DeepSeek, or Kimi for a few cents a session instead of renting a frontier lab by the month. - **Deep-agent planning + subagents + skills**, with `AGENTS.md` and MCP — the cross-vendor conventions Claude Code and Codex use too. - **A Textual TUI** — terminal-native, no web UI to babysit. That last part is the tell: dcode is built for a terminal that's always on, not a laptop you close. ## Why a real microVM for DeepAgents Code Here's the truth most "run an agent" guides skip. A coding agent runs `bash`, and dcode's planner runs whatever it decides — install this, rewrite that, fetch the other thing. On your laptop, that's the whole machine. A shared container shares the host kernel, with 200-plus known weaknesses in the stack — the wrong boundary for an agent acting on its own. **jurniti gives DeepAgents Code its own Firecracker microVM** — its own kernel, a KVM hardware boundary, one tenant per box. The same isolation serverless platforms use. Give dcode root inside its own VM and a runaway command, a bad install, or a poisoned dependency all stay contained to that box. Reprovision and you're clean in seconds. **Your model relationship stays yours.** jurniti's [BYOK](/pricing) is architectural: drop an `OPENROUTER_API_KEY` (or any provider key) and the credential lives only in your VM. Traffic leaves straight for the provider, and jurniti is never in the path. No proxy, no token markup — so the cents-per-session math on open weights stays cents. dcode runs on **Starter (1 vCPU / 2 GiB)** for most work. And it's reversible: a **30-day money-back guarantee**, workspace kept 7 days after cancellation. ## Run DeepAgents Code in 3 steps By hand this is a VPS, the LangChain installer, Python plumbing, auth, a tmux setup for durable sessions, and patching for as long as you run it. jurniti runs the box; you keep the agent. ### 1. Pick a plan and pay [Starter](/pricing) is enough to start. Check out, and the provision chain runs automatically the moment payment confirms. ### 2. Your microVM boots with dcode ready A fresh Ubuntu microVM boots with DeepAgents Code pre-installed and on your PATH. You get an in-browser terminal — nothing to SSH into. Wire a model and go: ```bash export OPENROUTER_API_KEY=... dcode --model openrouter:z-ai/glm-5.2 # open weights, cents per session # or point it anywhere: dcode --model anthropic:claude-... / openai:gpt-... ``` ### 3. Put it to work Hand dcode a task and let it plan, spawn subagents, and run. Your `AGENTS.md`, skills, and config live on the persistent home, so they survive every restart. ## What people run DeepAgents Code for - **Cheap open-weight coding.** GLM-5.2 or DeepSeek doing real work for cents, not a monthly seat. - **Model bake-offs.** Swap `provider:model` and run the same task across labs — same workspace, same agent. - **Pinned environments.** Your `AGENTS.md`, skills, and MCP servers — set up once, always there. - **Walk-away runs.** Plan it, close the laptop, come back to the result on a box that never sleeps. Snapshot a configured workspace into a [template](/templates) and fork it for the next project — your model wiring, never your credentials (scrubbed on publish). ## jurniti vs a laptop vs a DIY VPS | | jurniti | Your laptop | DIY VPS | |---|---|---|---| | Always on | Yes | No — it sleeps | Yes | | Isolation for an autonomous agent | Firecracker microVM (KVM) | Your whole machine | Shared kernel | | Model key | Yours, never leaves the VM | Yours | Yours | | Blast radius of a bad command | One disposable microVM | Your whole machine | Your whole server | | Setup | Automatic · 124ms boot | Local install | Manual | | Refund | 30-day money-back | n/a | None | ## Pricing Starter is **$25/mo** ($250/year, two months free) and runs DeepAgents Code for most work. Pro is **$49/mo** and Max **$99/mo** give heavier runs more room or add a custom TLS subdomain. Every plan includes the 30-day money-back guarantee, and your model spend is separate and never marked up — which, on open weights through OpenRouter, is the cheapest serious coding agent you can run. Running a different agent? See the guides for [Claude Code](/blog/claude-code-hosting), [OpenAI Codex CLI](/blog/codex-cli-hosting), and [OpenCode](/blog/opencode-hosting), or compare plans on the [pricing page](/pricing). ## Frequently asked questions ### What is DeepAgents Code? DeepAgents Code (the dcode binary) is LangChain's terminal-first coding agent, shipped as the deepagents-code PyPI package. It's a Textual TUI built around LangChain's deep-agents architecture — planning, subagents, skills, and AGENTS.md — and it's truly model-agnostic: you point it at any provider:model string (OpenRouter, Anthropic, OpenAI, Fireworks, Baseten, or a local model) instead of being locked to one lab. ### Can I run DeepAgents Code on a server? Yes. dcode is terminal-native, so an always-on box is its natural home. jurniti runs DeepAgents Code on a dedicated Firecracker microVM with an in-browser terminal and a persistent home — wire your model once, close your laptop, and the agent keeps your config, skills, and AGENTS.md across every restart. ### Which models can I use with DeepAgents Code? Any of them. dcode takes a provider:model string, so you can run open weights like GLM-5.2, DeepSeek, or Kimi through OpenRouter for cents — no $20/mo frontier subscription required — or point it straight at Anthropic, OpenAI, Fireworks, or a local endpoint. jurniti is bring-your-own-key: the credential lives in your VM, traffic goes straight to the provider, and we never see your tokens or mark up usage. ### Why run DeepAgents Code on a microVM? dcode runs real shell commands its planner decided on — code no human reviewed. On your laptop that has the run of your machine. On jurniti, dcode gets root inside its own Firecracker microVM with its own kernel and a KVM hardware boundary, one tenant per box. A runaway command or a poisoned dependency stays contained to one disposable VM you can reprovision in seconds. ### How much does it cost to host DeepAgents Code? Plans start at $25/mo ($250/year — two months free), and every plan has a 30-day money-back guarantee. dcode runs on Starter (1 vCPU / 2 GiB) for most work. Your model spend is separate and never marked up, because you bring your own key — and with open weights through OpenRouter that bill is often pocket change. ### Is this an official LangChain product? No. jurniti is independent managed hosting for DeepAgents Code. We install the upstream deepagents-code package at first boot — no fork — so you run the real dcode and authenticate with your own provider keys.