Local & self-hosted appsSolo Systems (open source; creator Jehuang / GitHub @1jehuang)
jcode
Decision matrix
jcode app decision matrix
Overview
jcode is a free, MIT-licensed terminal coding agent written in Rust, with no subscription of its own: it connects to Claude, OpenAI, Gemini, local Ollama or LM Studio models, or an OpenRouter-style aggregator, and you pay only that provider. Pick it for a lightweight agent you can run against a model you already have access to. It's new and fast-moving, so treat every session like a first trial.
Other good fits
Anyone who wants to try a coding agent's diffs and commands from the terminal without a monthly subscription.
+ Pros
Read pros
jcode is an open-source (MIT) terminal coding agent built in Rust by an independent developer. It has no subscription of its own: you connect it to a provider (Claude, OpenAI, GitHub Copilot, Gemini, Azure OpenAI, Alibaba Cloud through native OAuth; Ollama or LM Studio for local models; OpenRouter, Groq, Mistral, Perplexity, Together AI or DeepSeek as aggregators; or any OpenAI-compatible endpoint), and you pay that provider directly, or nothing at all with a local model.
Pick it when…
You already have API access or a subscription to a model provider and want a free agent to drive it from the terminal.
You run models locally with Ollama or LM Studio and want an agent harness with no per-seat cost.
You want to try several providers from one tool instead of installing a separate client for each.
− Cons
Read cons
An agent producing a plausible diff doesn't mean the change is correct, secure or compatible with the rest of your codebase; your own tests and review still matter. Its multi-agent "swarm" mode, browser automation and memory features add moving parts with real system and credential access, so understand what a feature does before turning it on. Being new and fast-moving cuts both ways: features and defaults can change between versions, and it hasn't had years of real-world hardening the way established tools have.
Pick something else when…
You want one bill from one vendor for both the agent and the model.Claude Code or Codex bundle the agent and a specific model family together.
Your team is already inside GitHub's workflow.GitHub Copilot puts the agent in pull requests and issues where your team already works.
You want a more established terminal agent with a longer track record.Aider has been around longer and has broader documentation and community troubleshooting.
Avoid if
You need a mature, widely adopted tool with a long track record; jcode is new and changes quickly.
Workflow
Install from the documented script or Homebrew tap → connect one provider you're comfortable testing with → open a disposable repository → ask for one narrow change → read the plan, the commands it wants to run, and the diff before accepting → run the repository's own tests → keep or revert.
Cost
jcode itself is free and open source. You pay only your chosen model provider: nothing extra for a local Ollama or LM Studio model, or standard API token pricing for a cloud provider like Claude or OpenAI. There's no jcode-specific plan or credit system to track.
Running effort is mostly the same as any coding agent: reviewing diffs, watching for a runaway loop of tool calls burning API tokens, and keeping API keys out of shared transcripts or config files committed to a repo. Because the project moves fast, expect to re-check its documentation each time you upgrade.
Restrictions
Installation scripts and provider connections have real system and credential consequences: review what the install script does before running it, and scope API keys narrowly. jcode's documentation doesn't establish a safe default permission policy for every environment, so set that up yourself before pointing it at anything beyond a disposable repository.
First project
Documentation-based suggestion — not run here.
Read first project
Ask it to fix a small, obvious issue in a throwaway repository.
Install with curl -fsSL https://jcode.sh/install | bash (or the Homebrew tap on macOS).
Connect a provider you already use, for example an existing Claude or OpenAI API key, or a local Ollama model if you'd rather spend nothing.
In a disposable Git repository, ask jcode to fix one small thing, such as a typo in a README or an obviously wrong comment.
Read the diff and the commands it ran before accepting anything.
Run the repository's own tests or linter yourself; don't take the agent's word for correctness.
Done when the diff touches only the lines you expected and your own test or lint run passes.
Links & evidence
Documentation-based · not hands-on · Sep 23, 2026
Rests on jcode's GitHub repository and its jcode.sh site and docs, checked 23 September 2026. jcode was not installed and no agent session was run for this review.
jcode repository · Sep 23, 2026 — Installation methods, supported providers (OAuth, local, aggregators), MIT license, memory and swarm features.
jcode site · Sep 23, 2026 — Product homepage confirming the terminal harness and its free, bring-your-own-key model.
Work with a coding agent from a terminal-based Rust harness.
Connect cloud subscriptions, API keys, or local Ollama and LM Studio models.
Use memory, background tasks, and parallel agents across projects.
Solo Systems (open source; creator Jehuang / GitHub @1jehuang) · byok · self-host
terminal coding
local model
persistent memory
swarm coordination
side panels
background tasks
open-source
write-code
automate
View sourced imageOfficial jcode README screenshot of the terminal provider picker. Provider availability and labels reflect the captured version.
jcode — 1jehuang/jcode README
Owner-approved use of public official interface screenshots, as recorded in public-visuals-batch-1.json, 13 September 2026. Copyright retained by Jehuang and jcode contributors; attribution retained. Editorial product illustration; no open licence asserted. Converted on 15 October 2026 to a 1600px-wide WebP (quality 80) for faster page loads; the original image remains in the media library.
