hexiel is an AI coding agent for your terminal and browser. It reads your project,
runs your tests, fixes bugs and builds features — on local models running on hardware you
already own, or on any big cloud model when you want one.
AI coding tools are amazing — and most of them are a monthly bill wired to someone else's
servers. hexiel gives you the same kind of agent, but you own it, you choose the model, and it's designed to be
fast on the computer you already have.
⌂ Local-first
Runs on Ollama, vLLM or LM Studio on your own machine or home server. Your code never has to leave your
network. Prefer a frontier model for a hard problem? Switch to Claude, GPT, Gemini or OpenRouter in one
command.
⚡ Fast on modest hardware
Anything that doesn't need a model — searching, mapping the project, waiting on tests, parsing logs —
runs locally in Python. A lean prompt halved the time to process each request on an Intel iGPU mini-PC
(46.5 s → 23.7 s).
♥ Free, and staying that way
The agent is open source under GPLv3 with the Commons Clause. Use it, change it, share it, contribute back — free. (Only the optional Remote add-on is paid.)
It just can't be repackaged and sold without permission, so no one can turn it into the subscription
it was built to replace.
Everything you'd expect. Plus what you'd wish for.
Two interfaces, one agent
A fast full-screen terminal UI and a browser UI share the same brain — plus one-shot headless runs
for scripts and an OpenAI-compatible API, so tools like OpenWebUI can talk to it.
Type / for a command menu with descriptions
A live activity line shows what it's doing, how long, how many tokens
Edits and commands ask first — or go hands-free with /auto
Remembers, plans and picks up where it left off
Reads your project's AGENTS.md / CLAUDE.md, keeps file-based memory, and
tracks multi-step work in a todo.md that it maintains automatically — so if your
laptop dies mid-task, hexiel -c resumes from an accurate checkpoint.
Auto-compaction at a Claude-Code-sized context window
Durable facts are saved to memory automatically
Each project keeps its own history and state
Every model you can use — right now
Press Ctrl+P: hexiel asks every provider you're connected to what's available at that
moment — your local Ollama, Ollama Cloud, Claude, OpenAI, OpenRouter — and lets you search and switch.
Never a stale list.
You're in control of every setting
Open /config (or ⚙ Settings in the browser), click a value, pick from the real options —
including the models your server actually has — and choose this session or
save permanently.
Edits one line of your config and keeps your comments
Updates itself with /update and resumes your conversation
Sees screenshots
Type @shot.png and vision models see it.
Knows your toolchain
Detects pytest, vitest, Go and Cargo test runners plus Docker and Podman — and runs any other build tool through the shell.
Researches the web
Search + page reading built in, no API key needed.
Answers with graphics
Ask about the weather and hexiel shows a live forecast card beside the answer — hourly, 4-day, rain chances. No API keys, no sign-ups.
Brings your plugins
Claude Code skills and OpenWebUI tools work unchanged.
Fixes itself — for everyone
Hit a bug in hexiel? /heal fixes it in a safe copy; review the diff, keep it, or send it back so every user gets the fix.
Merge requests for your code
/ship branches, commits and opens a GitLab MR for your project's changes.
Batches in Python
One local script instead of ten slow model round-trips.
Cross-platform
Linux, macOS and Windows (PowerShell built in).
Knows your versions (RAG)
Add the docs for Java 21, Tomcat 10.1, Angular… and hexiel answers from them, with citations. Free and local.
Speaks MCP
Local and remote MCP servers; your Claude Code .mcp.json works as-is.
Every major model, preset
/model add sonnet, gemini, deepseek, ollama-qwen… one command.
Private by default
Edits and commands ask first; local models keep code on your network.
Every model, live
Ctrl+P lists every model you can use right now — local, Ollama Cloud, Claude, GPT… never stale.
Never stalls on limits
Hit a rate limit? It counts down and resumes — or continues on your local model.
Undo & plan mode
/undo reverts a turn's edits; /plan proposes changes before making any.
Multi-line prompts with a clean, sanitized paste. Copy an image and press Ctrl+V (or paste in the browser) to attach it; long pastes collapse to a tag and expand when you send. Terminal clipboard images need wl-clipboard/xclip (Linux) or pngpaste (macOS).
Shrugs off network blips
A dropped or timed-out model connection is retried automatically. If it breaks mid-reply you're told, and nothing half-finished is saved.
Cloud brains for the plan. Your own model for the work.
Run a free local model for almost everything and spend cloud tokens only where they count.
For a big task, a strong cloud model writes a plan of small, self-contained steps; your local model then does each one
in a fresh, short context and checks its own work before moving on.
📐 Sized up for you
hexiel judges each request — cheap rules first, your local model only for toss-ups — and offers to delegate the big ones.
Or type /delegate yourself.
✓ Every step is checked
Each step ends with a test, build or search that has to pass. Failures are retried, then handed to the planner once, then
reported plainly. You approve the plan before anything changes.
⇄ Pick the models in the app
Choose the planner and the worker in /config or the browser settings — any mix of local, Ollama cloud, Claude or
OpenAI-compatible models. Stopped plans resume with /delegate resume.
hexiel stays free. Hexiel Remote is an optional add-on for when you want your agent
to reach beyond your desk — control the hexiel on your own machine from a browser or phone, and let it
work while you're away.
⇄ Remote control
Run hexiel remote login once, then type /remote-control in any chat. Your machine connects
out to the relay — nothing is opened on your network. Sign in at remote.hexiel.tech from any browser
or phone, pick the chat (up to 3 at a time) and you're in that same conversation, with the same approvals before edits and commands
— and a Stop button that works from anywhere.
⏱ Automations
Schedule a task ("every Monday 8am, summarise last week's sales") or give it a webhook URL
that CI, a form or a monitor can call. Runs execute on your machine against your files and
model, are read-only unless you opt a schedule into full access, and results go to your account page or
a Slack or Discord webhook.
🔒 Built to be safe
Sign-in plus an active subscription on every connection, revocable per-machine tokens, webhook payloads
handed to the model as untrusted data, and a password check before any automation is created.
Traffic passes through the relay over TLS (it isn't end-to-end encrypted).
$ $8 / month
14-day free trial — a card is taken up front and billed when the trial ends unless you cancel.
Payments by Stripe; manage or cancel any time from your account page.
Researchers, analysts and small-business owners: ask in plain English. hexiel does
the work locally — the AI only sees summaries, so it's fast on a small computer and cheap on a cloud model.
▦ Spreadsheets & data
"Which region beat its target — sales.csv vs targets.xlsx?" hexiel answers with real SQL over your
CSV, Excel and Parquet files and edits Excel without breaking your formatting.
⛁ Your own database
"Keep a list of my customers and invoices." hexiel creates and maintains a private local
database for you — no servers, no setup — or connects to Firebase and Supabase.
⏰ Jobs on a schedule
"Every Monday at 8am, summarise last week's sales and flag anything odd." It runs even when hexiel
is closed — on Linux, macOS and Windows — and the result lands in your inbox.
◔ Charts in seconds
"Chart revenue by month." You get an interactive chart that opens in any browser.
⌕ Research with sources
Ask a big question: hexiel reads many pages at once and writes a report with numbered sources.
Or drop in your PDFs and Word files and get answers that cite the exact page.
⚙ Connect your tools
One command adds Firebase, Supabase, Notion, GitHub or a real browser — plus any MCP server you already use.
How hexiel compares
Claude Code, opencode and Aider are excellent tools — we use and respect them. Here's an
honest look at where hexiel fits. Facts checked against each project's official docs on
(pricing and the remote-control row re-checked ).
hexiel compared with Claude Code, opencode and Aider
Feature
hexiel
Claude Code
opencode
Aider
Open source
YesGPLv3 + Commons Clause
Noproprietary
YesMIT
YesApache-2.0
Price of the tool
Yesfree — pay only for cloud models you choose
PaidPro/Max/Team/Enterprise plan or pay-per-token API
Yesfree; optional paid model plans
Yesfree; you pay your LLM provider
Local models (Ollama, vLLM, LM Studio…)
Yestuned for modest hardware
NoClaude models only (officially)
Yes
Yes
Terminal UI
Yesfull-screen
Yes
Yes
Partialchat REPL, not full-screen
Browser / desktop UI
Yesbuilt-in browser UI
Yesweb, desktop and IDE apps
Yesweb UI; desktop app
Partialexperimental --gui
Headless / scripting
Yes
Yes
Yes
Yes
Project memory + auto-compaction
Yesplus auto-maintained todo.md and memories
Yes
Yes
Partialconventions file via config
Uses Claude Code skills
Yesreads ~/.claude/skills
Yesnative
Yesreads .claude/skills
No
Screenshot / image input
Yes
Yes
Yes
Yes
Remote control from a phone / browser
Paid add-onHexiel Remote, $8/mo, 14-day trial
YesRemote Control, on Pro/Max/Team/Enterprise (not API keys)
Measured on an Intel iGPU mini-PC, not a datacenter. Searching, mapping, waiting on tests and parsing
run in Python; a python tool batches many steps into one local script; the prompt is kept lean.
Thinking can be switched off per model for speed.
Upkeep it does for you
Multi-step requests become tracked tasks with a checkpoint refreshed after every step, and durable facts
are saved to memory during compaction — in Python, at zero token cost.
Fits into OpenWebUI
Serve hexiel as an OpenAI-compatible model that OpenWebUI (or any OpenAI client) can connect to, and run
existing OpenWebUI tool, pipe and filter plugins inside hexiel without porting them.
Documentation for the versions you actually run
Models guess at which Java, Angular or Tomcat you use. hexiel knowledge add <docs-url> -t java-21
downloads the official docs (or copy your own PDFs and Markdown into a folder), and
hexiel knowledge sync indexes whatever is new or changed. Search combines keywords with local
Ollama embeddings, so answers come from your versions’ docs and cite them. Free, open source and
fully local. The first sync is slow on modest hardware (a 40-page site takes a few minutes); later ones
only embed what changed. How it works.
A small model for the small jobs
Map read, grep and web_fetch to a small, fast model — even a
Raspberry Pi — and big outputs are condensed for your request before your main model reads them. It
keeps their context clean and queue free, never touches writes or edits, and falls back to the raw text if the
helper is slow or down. hexiel also tells you when another window is using the same local model.
How it works, with Pi measurements.
See what you attach
Paste, drop or mention an image or PDF and it appears — thumbnails and a PDF reader in the browser UI,
clickable lines (and inline thumbnails with pipx install 'hexiel[view]') in the terminal.
Take over a chat from your phone
Type /remote-control in a chat you already have open and that same conversation is drivable from
any browser through Hexiel Remote; it asks you to sign in first if needed.
Fixes and ships through merge requests
Crashes are recorded as incidents; /heal reproduces, fixes, tests and proposes a merge
request, and /ship turns your work into one. Nothing merges without a human.
Which one is right for you?
If you live on Claude models and want deep IDE and desktop integration, Claude Code is
superb. If you want a permissive MIT license, opencode is a great choice. If you want a lightweight git-centric pair programmer, Aider is proven. If you want
an agent that is free, runs well on modest local hardware, maintains its own memory and todo, and plugs
into OpenWebUI — that's what we built hexiel for.
Try it in two minutes
You need Python 3.11+ and a model: Ollama for local, or an
API key for a cloud provider.
pipx install 'hexiel[serve]' # no pipx? see the note below
ollama pull qwen3.6:35b-a3b-coding # or use a cloud API key
cd ~/your-project && hexiel
winget install Microsoft.PowerShell
pipx install "hexiel[serve]" # or: pip install "hexiel[serve]"
ollama pull qwen3.6:35b-a3b-coding # or use a cloud API key
cd your-project; hexiel
Don't have pipx, or does pip install say
externally-managed-environment? Modern Linux and Homebrew Python refuse system-wide installs on purpose.
Install pipx first (Arch: sudo pacman -S python-pipx · Debian/Ubuntu: sudo apt install pipx
· Fedora: sudo dnf install pipx · macOS: brew install pipx), then run pipx ensurepath.
Or use a plain virtual environment:
python -m venv ~/.hexiel-venv && ~/.hexiel-venv/bin/pip install 'hexiel[serve]'.
Want to hack on hexiel itself? Install from source.
Run hexiel --doctor any time for preflight checks with copy-paste fixes, or
hexiel --serve for the browser UI.
Who we are, and why hexiel exists
hexiel is designed and built by JKagiDesigns LLC,
a software studio in Kingston, Massachusetts. Our day job is building custom business software for small
companies — software that replaces the subscriptions that cost too much or do too little.
We wanted an AI coding agent that worked the same way: one you own. One that runs on the mini-PC
in the closet instead of a monthly bill, that doesn't waste tokens watching a test suite scroll by, and that
you can open up and improve. So we built hexiel — a guardian angel for your codebase — and we're giving it
away.
The name nods to that idea of a watchful helper; the logo is a pair of wings around an H,
under a halo.
Use hexiel free, at home or at work, on any project
Read, study and modify the source code
Share it and fork it
Contribute improvements back through a merge request
You can't
Sell hexiel, or a product or service whose value comes from it, without permission from JKagiDesigns LLC
License: GPLv3 with the Commons Clause.
Want to build something commercial on top of it? Talk to us.
Questions
Is hexiel really free?
Yes. The tool costs nothing and has no subscription. The only paid piece is the optional Hexiel Remote add-on ($8/month, 14-day trial) for remote control and cloud automations. If you use a cloud model you pay that provider for
what you use; local models are free.
Can it run completely offline / locally?
Yes — point it at Ollama or any OpenAI-compatible server on your machine or LAN. Web search obviously
needs the internet, but everything else works locally.
Can I use a cloud model only for the hard parts?
Yes — that is what delegation is for. Set a planner model (say, an Ollama cloud model) and a worker model
(your local one). For big requests the planner writes a short plan once and your local model does the work, so cloud tokens go to
the thinking, not the typing.
Where are all the settings documented?
On the config file reference: every config.toml setting with its default, model profile examples,
permission rules, MCP servers, environment variables and fixes for common mistakes.
What hardware do I need for local models?
Our test machine — an Intel mini-PC with integrated graphics — runs a 35B mixture-of-experts coding
model at about 12 tokens/second. That model needs roughly 25 GB of memory, so plan for 32 GB of RAM or more.
Smaller machines can run smaller models, or use a cloud model.
Which local model should I use?
Mixture-of-experts models with ~3B active parameters that support tool calling and vision —
qwen3.6:35b-a3b-coding passed all our checks. Set reasoning_effort = "none" on slow
hardware so it doesn't spend minutes "thinking".
Will my existing Claude Code skills work?
Yes. hexiel reads ~/.claude/skills directly, plus OpenWebUI tool/pipe/filter plugins.
Can I give hexiel the documentation for the versions I use?
Yes — it’s built in and free. hexiel knowledge find "Tomcat 10.1 documentation" searches the web
for the official docs, hexiel knowledge add <url> -t tomcat-10.1 downloads a docs site, or you copy
files (Markdown, HTML, PDF, Word) into a topic folder yourself. hexiel knowledge sync then
indexes what changed. Models search it automatically and cite the page. It all runs locally; semantic search
uses Ollama’s nomic-embed-text, and keyword search works without it.
Does hexiel support MCP?
Yes — local (stdio) and remote (HTTP) MCP servers. It reads Claude Code's .mcp.json format,
so existing setups work unchanged, and you can expose only some of a server's tools to keep prompts
small on modest hardware. /mcp shows what's connected.
What happens when hexiel finds a bug in itself?
Run /heal (optionally describing the bug). hexiel fixes and tests it in an isolated copy of
its own source — never in your project — then shows you every changed file with a diff. Untick anything
you don't want, then send it back as a merge request so every user benefits, or
keep it just for yourself.
How fast can I connect my favourite model?
One command: /model add sonnet, gpt, gemini, deepseek,
kimi, openrouter, ollama-qwen and more. hexiel --presets
lists them all, and MODELS.md
has copy-paste config for each.
I'm not a programmer — can I still use it?
Yes. Ask about your spreadsheets ("which product sold best last quarter?"), have it keep a small
database for you, make charts, research a topic with sources, or schedule a weekly report. It works in
the browser too: hexiel --serve.
Is it safe to let an AI run commands?
By default every edit and command asks for your approval first. The browser UI is local-only unless you
explicitly expose it, and it warns you loudly if you do.
How do I contribute?
Fork, branch, open a merge request — or let hexiel do it with /ship. Changes are reviewed
and approved by the maintainers.
