codecraft/README.md

CodeCraft

A self-hosted coding agent for your browser and terminal: any model, your tools, your rules.
Run Claude, Gemini, OpenRouter, opencode or your own models as coding agents with files, shell, MCP and subagents,
approve what they do from any device, automate recurring work, and see exactly what every request costs.

Install · How it works · Modes · Models · Terminal client · Automations · Integrations · Users · Statistics · Configuration · Themes · Development

_{A session: the agent works through its task list, shows diffs, and asks before anything outside the rules.}

Any model Anthropic natively, plus everything that speaks the OpenAI API: Gemini, OpenRouter, opencode, fal.ai, Ollama, vLLM, llama.cpp. Mix them in one chat, with automatic fallback.

Real tools Files, shell, web, an interactive browser, LSP diagnostics, task lists, subagents, skills – and every tool of your MCP servers.

You decide Permission cards with diffs, questions from agents and MCP servers, per-tool rules. On your phone, with push notifications.

Every token counted Tokens, cache hits, prefill and output speed, and cost per request – per chat, per day and per model.

Install

You need Docker with Compose. The image is published for amd64 and arm64.

mkdir codecraft && cd codecraft
curl -fsSLO https://git.familie-zink.org/zink/codecraft/raw/branch/main/docker-compose.release.yml
curl -fsSL --create-dirs -o docker/seccomp-sandbox.json \
  https://git.familie-zink.org/zink/codecraft/raw/branch/main/docker/seccomp-sandbox.json
curl -fsSL -o .env https://git.familie-zink.org/zink/codecraft/raw/branch/main/.env.example
$EDITOR .env                    # workspace path, port, user id, time zone
docker compose -f docker-compose.release.yml up -d
docker compose -f docker-compose.release.yml logs codecraft   # one-time setup code for the first admin

Open http://localhost:8787, enter the setup code and create the admin account. Then add a provider under Settings → Providers & models. .env only holds deployment settings; keys, tokens and everything else are set in the UI and stored in the configuration file (${NAME} references to environment variables still work).

Update: docker compose -f docker-compose.release.yml pull && docker compose -f docker-compose.release.yml up -d. Releases are named vYYYYMM-NNNNN (a counter per month). Pin one with CODECRAFT_VERSION=v202611-00001 in .env. Admins see a notice when a new release is out.

From source: git clone https://git.familie-zink.org/zink/codecraft.git && cd codecraft, cp .env.example .env, then docker compose up -d --build.

All state – configuration, database, backups, the agent's home directory – lives in the Docker volume codecraft-data; recreating the container keeps it. Your projects are mounted from WORKSPACE. The port is bound to 127.0.0.1: put a TLS reverse proxy in front of it (with WebSocket support) and set auth.public_url – installing the app and push notifications require HTTPS.

Moving to another host (for example Unraid with bind mounts instead of the Docker volume): scripts/move-server.sh root@newhost:/mnt/user/appdata/codecraft --workspace /mnt/user/projects copies the workspace, stops CodeCraft, streams the data volume to <dir>/data, writes a docker-compose.yml (and a start.sh for hosts without Compose) with bind mounts, starts the published image there and runs scripts/smoke-test.sh. The old container is only stopped, so you can roll back. --help lists the options (owner, port, bind address, dry run). --tools-dir /mnt/disks/ssd/codecraft puts everything agents install or download – the agent home (/data/home: toolchains, pip/npm/cargo packages, caches), the users' homes and /tmp – on a separate disk; the data folder keeps only the database, configuration and backups. Step by step, with rollback and troubleshooting: docs/MOVING.md.

How it works

CodeCraft runs on your server, next to your projects. You talk to it from a browser, the installed app or the terminal client; the agent keeps working when you close the tab.

1. Pick a folder and a task. The start screen asks what you want to do – build, plan, craft something, model a part for 3D printing – and where: a new folder, an existing one or a git repository.
2. Say what you want. The agent plans with a task list, reads the code and makes changes.
3. Approve or answer. Anything outside your rules shows up as a card: allow once, allow for the session, or reject with feedback. Questions from agents and MCP servers too.
4. Delegate. The agent hands work to subagents – in parallel or in the background, on other models and hosts – and continues while they run.

Every change can be reviewed hunk by hunk, rewound to a checkpoint, or isolated in a git worktree. Sessions can be pinned, archived and filed into folders; Ctrl+K opens the command palette, and prompt templates with placeholders save typing.

Mobile: answer questions and approve changes from your phone. Share links and text from other apps into a chat.

Settings: providers, models, prices, MCP, permissions, users, templates and backups – all in the UI.

Modes

Mode	What it does
build	Full access: implement, edit, run.
plan	Read-only: investigate, ask, and propose a step-by-step plan. One click switches to build.
craft	Study a target – a website, app, game or old program – and build your own version, check it visually and refine. Websites are captured at several screen widths in a headless browser and compared with the build. Clean room uses only what a visitor sees; deep mode may read the source.
custom	Your own modes and subagents – in the UI, the config file or as Markdown in .codecraft/agents / .claude/agents.

Projects keep working the way they do in Claude Code and opencode: AGENTS.md, CLAUDE.md, slash commands in .claude/commands and skills in .claude/skills are picked up automatically. The agent keeps notes about a project in .codecraft/memory.md; your personal memory applies to every project.

Models

Every provider is an endpoint with a list of models. Each model can have a context size, output limit, reasoning effort, extra request options and prices per million tokens.

• Several models per chat: switch the model at any time, give modes a fixed model, and run subagents on other models – e.g. a coding model on one GPU host and a small model that writes documentation in the background on another.
• Fallback: if a host is unreachable or its budget is used up, the request goes to the model's fallback.
• Concurrency limits: max_concurrency per host or model queues requests for servers that process one at a time; waiting time is shown and never counted as model latency.
• Budget warnings: admins get a push notification when the total, a provider's or the image budget, or an opencode Go window reaches the warning share (default 80 %) and when it is used up; the status overlay shows all budgets with bars.
• Prices: Fetch prices fills in prices for OpenRouter and opencode (Zen and Go), including long-context tiers and peak-hour factors, which are applied per request.
• Status: every five minutes each provider's model list is fetched (no tokens used), so you see at a glance which hosts answer and which keys are rejected. Saving a provider checks it right away, and Test connection checks URL and key before you save.
• Images (fal.ai): the agent drafts cheap previews (FLUX schnell), asks which one to refine, and creates the final image with the best model for the purpose – Nano Banana Pro refines the chosen draft, Ideogram handles lettering, Recraft design work.

Terminal client

codecraft is a full-screen terminal client in the style of Claude Code and opencode. The agent loop, the tools and the chat history run on your machine, in the current folder. Signed in to a CodeCraft server, it uses the server's models, keys, budgets, modes, rules, skills, MCP tools and your theme, and the server counts usage and costs. Without a server, a guided setup connects Anthropic, OpenAI, Gemini, OpenRouter, opencode or a local model server (Ollama, LM Studio, llama.cpp, vLLM – detected automatically); both can be mixed.

curl -fsSL https://<your-codecraft-server>/cli/install.sh | sh   # Linux x86_64 / aarch64
codecraft login            # server URL, user name, password → creates an API token
codecraft                  # start in the current folder; /help lists the commands

The Docker image ships static binaries for Linux x86_64 and aarch64; the server offers them for download (Settings → Account) and codecraft update fetches new versions from it. Other platforms: cargo install --locked --git https://git.familie-zink.org/zink/codecraft codecraft-cli.

Automations

An automation is an agent that starts on a schedule. Its instructions are a Markdown file (.codecraft/jobs/<name>.md, only the prompt); name, schedule, mode, model and limits are set in the UI, with a graphical schedule editor (every few minutes, hourly, daily, on weekdays, monthly or a cron expression). New automation creates the file – blank with an example structure or from a template (dependency check, security review of recent commits, test suite health, daily summary, TODO report, documentation freshness) – and opens an editor with an assistant chat in which a model works out the instructions with you and writes the file. You can also ask an agent in any chat to set one up. Each run is a normal chat session, so you can read and continue it.

Webhooks start an automation from outside, e.g. on every push: the request is checked with a shared secret or a Gitea/GitHub HMAC signature, and its JSON payload is handed to the agent.

Integrations

• Telegram: connect your own bot (Settings → Account → Telegram) and use CodeCraft from Telegram – sessions, modes, models, approvals with buttons, photos and files; /help in the bot lists the commands.
• API tokens (Settings → Account) with full or read-only scope for scripts and the terminal client.
• Prometheus: set server.metrics_token and scrape GET /metrics with Authorization: Bearer <token> – requests, errors, tokens, cost and time per model, provider status, sessions, pending permissions.
• JSON logs: CODECRAFT_LOG_FORMAT=json writes one JSON object per line.
• Backups: daily at 03:00 into /data/backups (the last 30 are kept), optionally copied to any S3-compatible bucket (AWS, MinIO, Garage, Backblaze B2, Hetzner). Settings → Audit shows them, creates one on demand, downloads them and restores one – a stored backup or an uploaded zip, e.g. to move to another server. The current state is backed up first, then CodeCraft restarts with the restored database and configuration.
• Health: GET /api/health for monitoring, with version and commit.

Users and security

• Local accounts (Argon2 passwords) and single sign-on via OpenID Connect (Authentik, Authelia, Keycloak, …) with optional admin and access groups.
• Admins manage providers, MCP, permissions, users and shared folders and see all statistics. Users see their own sessions and statistics.
• Every user works in <workspace>/<username>; admins can share additional folders. Sessions are private and can be shared read-only. An audit log records sign-ins, permissions and configuration changes.
• Shell commands of non-admin users run in a bubblewrap sandbox: read-only system, only their folders writable, own home and processes, no server environment. It needs the security options in the compose file; without them non-admins cannot run shell commands.
• The browser tool reaches the internet and localhost, never addresses in private networks.

Statistics

Every model request is stored with input, output and cache tokens, time to first token, prefill and output speed, and its cost at the price configured at that moment. The dashboard shows totals, cost per day by model, tokens per day, an activity heatmap of the last 14 days, and tables per model, provider, mode, user and session – each chart also as a table. Turn on the daily cost summary to get a push notification every evening.

Configuration

One commented TOML file, /data/config.toml in the volume, created from config.example.toml on first start. The settings pages edit it in place and keep your comments. Secrets belong in the environment and are referenced as ${NAME}.

Section	Purpose
[server]	workspace, default and small model, currency, auto-compaction, subagent limit, metrics_token
[providers.<name>]	kind = "anthropic" | "openai", base_url, api_key, max_concurrency, models (with prices and fallback)
[mcp.<name>]	type = "stdio" (command, args, env) or "http" (url, headers)
[permissions]	allow / ask / deny per tool (globs like mcp__*); bash_rules for commands
[agents.<name>]	modes and subagents: model, tools, max_parallel, prompt
[auth], [auth.oidc]	sign-in, public URL, single sign-on
[sandbox]	shell sandbox for non-admins (auto, always, off)
[backup], [backup.s3]	daily backups, retention, S3 target
[images]	fal.ai key, budget, models with stage (preview / final), good_for and reference_param
[craft]	headless browser, screen widths, clean room

Compound shell commands (&&, ;, |) are only auto-allowed when every part is allowed.

Themes

Pick a base theme in Settings → General and switch light / dark / auto there or with the theme button in the sidebar (auto follows the system preference). Every base theme has a dark and a light variant. The palettes come from zink/trilium-themes (MIT, see web/src/themes/LICENSE-trilium-themes), which in turn are based on:

Theme	Variants	Palette source	License
Gruvbox	Dark, Light	morhetz/gruvbox	MIT
Ash	Dark, Light	original, Gruvbox-derived (zink/trilium-themes)	MIT
Harbor	Dark, Light	HANCORE-linux/omarchy-harbordark-theme	MIT
Coffee	Dark, Latte	megabyte0x/omarchy-coffee-theme	MIT
Solarized	Dark, Light	Solarized by Ethan Schoonover	MIT
Tokyo Night	Night, Day	folke/tokyonight.nvim	Apache-2.0

Further standard themes (palettes re-created from the published colour specifications):

Theme	Variants	Palette source	License
Nord	Dark, Light (Snow Storm background)	nordtheme/nord	MIT
VS Code	Dark Modern, Light Modern	microsoft/vscode	MIT
GitHub	Dark, Light	primer/github-vscode-theme	MIT
Catppuccin	Mocha, Latte	catppuccin/catppuccin	MIT
One	One Dark, One Light	atom/atom	MIT
Dracula	Dracula, Alucard	dracula/dracula-theme	MIT
Rosé Pine	Main, Dawn	rose-pine/rose-pine-theme	MIT
Everforest	Dark, Light	sainnhe/everforest	MIT
Kanagawa	Wave, Lotus	rebelot/kanagawa.nvim	MIT

Colours, syntax highlighting and chart palettes come from CSS variables, one file per variant in web/src/themes. To add a base theme <id>:

1. Copy gruvbox-dark.css / gruvbox-light.css to <id>-dark.css / <id>-light.css and change the values on :root[data-theme="<id>-dark"] / "<id>-light"; note source and license in the header.
2. Import both in themes/index.ts and add the base theme to THEMES in web/src/theme.ts.

The variables cover surfaces (--bg0–--bg3), text (--fg0–--fg3, --muted), accents, diffs, syntax (--syntax-*) and the browser bar colour (--theme-color). The chart palettes (--series-1–8 categorical, --seq-1–6 sequential) live in themes/base.css per light/dark mode and are validated for colour-vision deficiencies.

Development

Requirements: Rust (edition 2024) and Node ≥ 20.19.

# backend – writes ./data/config.toml on first start
cd server && CODECRAFT_CONFIG=../data/config.toml cargo run

# frontend with hot reload, proxies /api to 127.0.0.1:8787
cd web && npm install && npm run dev

# terminal client
cargo run -p codecraft-cli

Tests: cargo test --workspace · type check: cd web && npx tsc --noEmit.

core/    shared crate: types, config, providers, permissions, prompts
server/  Rust (axum, tokio)
  runner.rs    agent loop: streaming, tools, permissions, subagents, compaction
  tools/       built-in tools (files, shell, web, browser, agent, craft)
  mcp.rs       MCP client (stdio, Streamable HTTP, elicitation)
  jobs.rs      automations and webhooks
  auth.rs      users, sessions, API tokens, OpenID Connect
cli/     terminal client (ratatui)
web/     React + Vite PWA
  src/components  chat, composer, cards, settings, statistics
  src/themes      CSS themes
  src/i18n        English (reference) and German

The Docker build cross-compiles the server and the terminal client with cargo-zigbuild, so a multi-arch image builds quickly on either architecture: docker buildx build --platform linux/amd64,linux/arm64 -t <registry>/codecraft:<version> --push .

Releases are named vYYYYMM-NNNNN: year and month, then a counter that starts at 00001 every month. Collect changes under ## [Unreleased] in the changelog (and under unreleased in web/src/whatsnew.ts), then scripts/release.sh picks the next number, renames both, commits, tags, builds and pushes the image (:vYYYYMM-NNNNN and :latest) and publishes the Gitea release (--dry-run shows the steps). The Cargo.toml versions are not used for releases.

Translations: copy web/src/i18n/en.ts, translate the values and register the language in web/src/i18n/index.ts – the type checker reports missing keys.

Status

Early, but in daily use. See the changelog. Planned next:

• Terminal client builds for macOS and Windows
• MCP OAuth, resources and prompts
• The OpenAI Responses API

License

MIT © Richard Zink. The theme palettes keep the licenses listed under Themes.