Your AI assistants each start from zero, every session, in every tool.
gald3r drops a persistent, file-based brain into your repo — tasks, bugs, plans,
and constraints that every AI tool reads and writes.
Plan in one tool. Code in another. Nothing is lost between them.
Docs · Changelog · All 38 platforms
Stop watching scrolling logs.
gald3r's next act puts your entire AI crew inside a living, walkable 3D world —
characters with names, faces, voices, and desks, performing your project's real work in real time.
Watch them huddle over a bug. Argue in the war room. Celebrate a release by the pool.
Then step in yourself — on screen or in VR — and run the room.
And it's bigger than code. The same stage is built for every kind of project gald3r runs —
podcasts and newscasts with an anchor desk, music and video productions, episodic shows, skits,
choose-your-own-adventure worlds — describe your set in plain words, drop in reference pictures,
and the creator studio builds the world, the cast, and the show around your project.
Any project type. Any theme. Yours. Broadcast it 24/7.
Built on the gald3r engine below — the shared brain becomes a shared world.
First look is coming to the stream. Watch this space.
The Go rewrite is out now — signed on Windows, macOS, and Linux. The version badge above always shows the newest release; every link below resolves to it.
| OS | Installer | Portable (all binaries) |
|---|---|---|
| Windows | gald3r-windows-x86_64.msi (signed) | gald3r-windows-x86_64.zip |
| macOS | gald3r-macos-arm64.pkg (signed) | gald3r-macos-arm64.tar.gz |
| Linux | gald3r-linux-x86_64.tar.gz (with install.sh) |
gald3r-linux-x86_64.tar.gz |
Authenticode-signed on Windows, Developer-ID codesigned + Apple-notarized on macOS, sha256 sidecars on every asset. These links always resolve to the newest release.
Companion apps — optional, installed separately:
| App | What it is | Get it |
|---|---|---|
| Throne | Desktop control center for gald3r projects (Windows/macOS/Linux) | gald3r install throne — fetches the latest signed build from gald3r_throne releases (minisign-verified) |
| Longship | Terminal UI for chat + swarm work, same engine | Built on the same engine — public distribution lands in an upcoming beta |
Everything — the framework, all 38 platform overlays, the task engine — ships in one binary and provisions from it. This repo is the product's front door; the binary does the rest.
You use Cursor for one thing and Claude Code for another. Maybe Copilot at work, Codex on a side project, a local model when you're offline. Each one opens with no idea what you decided yesterday, what's already broken, or what you told the last one never to touch.
So you re-explain. Every time. In every tool.
gald3r puts a .gald3r/ folder in your project — plain markdown files for tasks, bugs,
plans, constraints, and cross-project coordination. Every AI coding tool you use reads
and writes the same files:
- Tasks and bugs that live in your repo, in your git history — not in one tool's memory
- Constraints your AI agents must follow — user-authored, per-project law
- Plans and PRDs that survive restarts, tool swaps, and team changes
- A deterministic engine (
gald3r) that runs the boring parts with zero LLM calls
No accounts. No server. No database to operate. No Docker.
1. Get the binary — from gald3r_core releases:
| OS | Fastest path |
|---|---|
| Windows | Download gald3r-windows-x86_64.msi, run it, open a new terminal |
| macOS | Download gald3r-macos-arm64.pkg, double-click, follow the prompts |
| Linux | curl -LO .../releases/latest/download/gald3r-linux-x86_64.tar.gz && tar xzf gald3r-linux-x86_64.tar.gz && ./gald3r-*-linux-x86_64/install.sh |
Verify with gald3r --version. (Full per-OS instructions, checksums, and source builds:
docs.gald3r.ai/install.)
2. Put the brain in your project:
cd /path/to/your/project
gald3r setup # creates .gald3r/ and walks you through
gald3r platform install cursor # or: claude, codex, opencode, copilot, windsurf, ...3. Open your AI tool and go. /g-setup in Claude Code, @g-setup in Cursor —
the brain is live and every AI you point at the repo can see it.
4. Your first five minutes — everything works from any terminal, any tool:
gald3r status # where the project stands, at a glance
gald3r task add "Ship the login fix" --origin owner-verbatim \
--objective "Fix the redirect loop on login and prove it with a test" \
--why "Users bounce on every failed login" --impacted src/auth
gald3r task list # the board, priority-sorted
gald3r doctor # health check: config, PATH, platform overlaysTasks carry their full story by design — who asked, what for, why, and what it
touches — so an AI (or a teammate) reading the task three weeks from now doesn't
have to re-derive your intent. Bugs work the same way: gald3r bug add.
New releases ship frequently. The update ritual, in order:
gald3r install update # fetch + install the newest signed release
gald3r -v # confirm the new version took
gald3r platform install cursor --force # regenerate each AI tool's overlay
gald3r platform install claude --force # (--force overwrites the old files;
gald3r platform install codex --force # run one line per tool you use)
gald3r upgrade-project # migrate .gald3r/ to the new schema
gald3r doctor # verify everything landed clean
gald3r task-sync-check -fix # reconcile any index/file driftThe platform overlays are generated from the binary, so re-running
platform install --force after every update keeps your AI tools' commands and
skills exactly in sync with the engine version you're running.
gald3r installs the same .gald3r/ task board, bug tracker, and CRASH overlay into every platform below — one binary, gald3r platform install <name>, and the files that platform actually reads land in your project. Full capability breakdown: PLATFORM_SUPPORT.md.
- Aider — also searched as aider-chat, Aider AI
- Amp — also searched as Amp Code, Sourcegraph Amp
- AstrBot
- Augment Code — also searched as Auggie, Auggie CLI
- Claude Code — also searched as Claude, Anthropic Claude Code
- Cline
- Continue.dev — also searched as Continue
- Cursor — also searched as Cursor IDE
- Deep Code CLI — also searched as Deep Code, deepcode-cli
- Devin Desktop — also searched as Windsurf, Cascade IDE
- GitHub Copilot — also searched as Copilot, GitHub Copilot CLI
- Google Antigravity — also searched as Antigravity IDE, Antigravity
- Goose — also searched as Block Goose, Goose AI agent
- Hermes Agent — also searched as Nous Hermes Agent, Nous Research Hermes
- JetBrains Junie — also searched as Junie
- Kilo Code — also searched as KiloCode
- Kimi Code — also searched as Kimi Code CLI, Moonshot Kimi Code, kimi-cli
- Kiro — also searched as Kiro IDE, Amazon Kiro
- Kiro CLI — also searched as Amazon Q Developer CLI, Q Developer CLI
- Mistral Vibe CLI — also searched as mistral-vibe, Mistral Code
- OpenAI Codex — also searched as Codex CLI, Codex
- OpenClaw
- OpenCode — also searched as sst/opencode
- OpenHands — also searched as All Hands AI, OpenDevin
- Pi — also searched as badlogic/pi-mono, Pi coding agent
- Qoder — also searched as Alibaba Qoder
- Qwen Code — also searched as Alibaba Qwen Code
- Replit Agent — also searched as Replit
- Roo Code — also searched as Roo Cline
- SubQ Code — also searched as SubQ
- Tencent CodeBuddy Code — also searched as CodeBuddy
- TRAE — also searched as ByteDance TRAE, TRAE IDE
- Void — also searched as Void IDE
- Warp — also searched as Warp terminal, Warp Oz
- Xiaomi MiMo-Code — also searched as MiMo-Code, MiMo Code
- ZCode — also searched as Z.ai ZCode, GLM ZCode
- Zed — also searched as Zed Editor
gald3r doesn't just live inside Claude Code and Cursor — it can drive them headlessly for autonomous pipelines, and it talks straight to your model backend of choice, local or cloud:
| Backend | What it is | Docs |
|---|---|---|
| Claude Code | Anthropic's terminal coding agent — g-go/g-go-go drive it headlessly for autonomous coding |
Platform guide |
| Cursor | The cursor-agent CLI — same headless autopilot story, one flag away |
Platform guide |
| Ollama | Local model runtime — gald3r init-providers auto-discovers every model you've pulled |
Provider docs |
| LM Studio | Local model runtime with a GGUF library and its own local server | Provider docs |
| OpenRouter | One API key, every hosted model — openai_compat, no adapter needed |
Provider docs |
Unsloth Studio  |
Built-in unsloth provider id, bearer-token auth, localhost:8888/v1 by default — gald3r's first-supporter, day-one integration |
Provider docs |
Full local + cloud provider setup: Models & Providers.
gald3r autopilot loop claims work off your board, implements it, reviews it, and
repeats — unattended. Every role (coordinator / implementer / reviewer) takes an
explicit provider and model; nothing is silently defaulted. --budget caps
implementation starts, --lanes caps parallel implementers.
Cursor (all three roles on Cursor's hosted Grok):
gald3r autopilot loop --budget 8 --lanes 3 --coordinator-provider cursor-agent --coordinator-model cursor-grok-4.6-high-fast --implementer-provider cursor-agent --implementer-model cursor-grok-4.6-high-fast --reviewer-provider cursor-agent --reviewer-model cursor-grok-4.6-high-fastCursor slug caveat:
cursor-agentdoesn't validate model names — an unrecognized slug silently runs (and bills) against a fallback model. Probe a new slug first:cursor-agent -p "reply OK" --model <slug>.
Claude Code (all three roles on Anthropic):
gald3r autopilot loop --budget 8 --lanes 3 --coordinator-provider claude --coordinator-model claude-sonnet-5 --implementer-provider claude --implementer-model claude-sonnet-5 --reviewer-provider claude --reviewer-model claude-sonnet-5OpenRouter — Ox Alpha workers (1M-context stealth model, free tier; the coordinator role currently requires a CLI agent — expansion to every provider is on the roadmap):
gald3r autopilot loop --budget 8 --lanes 3 --coordinator-provider claude --coordinator-model claude-sonnet-5 --implementer-provider openrouter --implementer-model stealth/ox-alpha --reviewer-provider openrouter --reviewer-model stealth/ox-alphaOllama — Qwen 3.8 27B on your own GPU (4-bit fits a 24 GB card; workers cost $0):
gald3r autopilot loop --budget 8 --lanes 3 --coordinator-provider claude --coordinator-model claude-sonnet-5 --implementer-provider ollama --implementer-model qwen3.8:27b --reviewer-provider ollama --reviewer-model qwen3.8:27bOllama — Ornith 1.5 9B on your own GPU (small, fast, fine for light lanes):
gald3r autopilot loop --budget 8 --lanes 3 --coordinator-provider claude --coordinator-model claude-sonnet-5 --implementer-provider ollama --implementer-model ornith:9b --reviewer-provider ollama --reviewer-model ornith:9bUnsloth Studio — Qwen 3.8 27B implementing, DeepSeek-V4-Flash reviewing
(Unsloth's Dynamic 3.0 GGUFs and
DeepSeek-V4 — match the model ids to
what your Unsloth Studio instance lists; UNSLOTH_API_KEY required):
gald3r autopilot loop --budget 8 --lanes 3 --coordinator-provider claude --coordinator-model claude-sonnet-5 --implementer-provider unsloth --implementer-model Qwen3.8-27B-GGUF --reviewer-provider unsloth --reviewer-model DeepSeek-V4-Flash-GGUFMix freely: a hosted frontier coordinator with local $0 workers is the
cost-efficient default. Register your endpoints once in providers.yaml
(gald3r init-providers, then gald3r providers add) and every example above
routes through it.
One product family, five pieces. ● owns it · ◐ involved in it.
| What you see | In plain terms | gald3r_core (engine) | gald3r_throne (desktop) | world_tree (cloud) | Framework (CRASH) | Longship (agent) | Docs |
|---|---|---|---|---|---|---|---|
| Task, bug & plan tracking | A file-based brain in your repo every AI reads and writes | ● owns the verbs + database | ◐ displays it | ◐ syncs it | ◐ defines the format | docs | |
| Rules & constraint enforcement | Always-on behavioral guardrails + your own per-project law | ● owns the checker + constraint verbs |
● ships the rule packs | docs | |||
| 38 IDE/platform integrations | Cursor, Claude Code, Codex, Copilot & 34 more, from one binary | ● generates every overlay | ● the content itself | matrix | |||
| Autonomous pipelines | g-go / g-go-go — claim, implement, verify, review, unattended |
● owns the orchestration | ◐ defines the playbooks | ◐ execution plane | docs | ||
| Desktop workbench | A native GUI over the same brain and engine | ● owns it | throne | ||||
| Cross-project coordination | Projects ask/order/sync with each other (WPAC) | ◐ the client | ◐ the registry | ● the protocol | docs | ||
| Live coordination backbone (Valkyrie) | Agents message you and each other mid-run; the local brain syncs to the cloud | ◐ the gald3r valk connector |
◐ its inbox | ● the sync target | ◐ the ledger format | docs | |
| Cloud sync & identity | Multi-device sync and sign-in for teams | ◐ gald3r valk client |
● owns it | docs | |||
| Memory & knowledge vault | Scoped cross-session memory (project/user/client), stash & reload for long runs, Obsidian-compatible vault | ● owns it | ◐ sync (roadmap) | docs | |||
| Personality system | A startup team of Norse gods voices your agents — Thor on performance, Sindri on craft, Loki on creative breakage | ◐ activation layer | ◐ shows it | ● the persona packs | docs | ||
| Skill packs | Curated capability bundles per project type — install a pack, get its skills locked to it | ● owns skill-pack / skills-lock |
● the bundles | docs | |||
| Third-party plugins | Install plugins from any git repo; enable/disable per project | ● owns gald3r plugin |
◐ SKILL.md format | docs | |||
| Themed reports | Status, review, and backlog reports rendered in your chosen theme | ● the render engine | ◐ displays it | ● theme definitions | docs | ||
| Workspace-Control | One control project governs many repos, with guarded member boundaries | ● owns it | ◐ the manifest format | docs | |||
| Experiments framework | Hypotheses with measurable gates and failure autopsies, tracked like code | ● owns it | ◐ the experiment format | docs | |||
| Generated CRASH reference | The full user-facing catalog of every command, skill, rule, agent, and hook shipped | ● generates it from source | ● the corpus | docs | |||
| Multi-format output | The same reports as markdown, HTML, JSON, or TOON — for humans and agents | ● owns it | docs | ||||
| QA, review & security gates | Deterministic validation that blocks bad state before it lands | ● owns the gates | ◐ the hooks that call them | docs | |||
| Release & code signing | One tag → signed binaries for all three OSes | ● owns the pipeline | docs | ||||
| MCP server | Other AI tools query your gald3r brain as MCP tools | ● owns it | docs |
gald3r_core, gald3r_throne, world_tree, and gald3r_longship ship as compiled
binaries (the first two public, the rest private) — this repo carries the framework
they all serve, plus everything below.
Marketable, on the way, and labeled as such:
| In the forge | What it becomes | Where it lives today |
|---|---|---|
| Marketplace | A curated, signed plugin registry on world_tree | Design (blocked on the cloud registry); git-installed plugins work today |
| Custom themes | A visual editor for your own report themes | Theme definitions ship; editor in progress |
| Suite installer | One manifest install: core + longship + throne together | Design complete; implementation next |
| Scoped memory sync | Your memory follows you across machines | Local memory shipped; cloud sync on the roadmap |
This repo is gald3r's front door — a landing page, not an install dependency:
| Path | What it is |
|---|---|
README.md |
This page |
PLATFORM_SUPPORT.md |
The full 38-platform support matrix — with the honest gaps marked |
platforms/ |
Per-platform reference docs — how each tool's project structure and extension surfaces actually work |
CHANGELOG.md |
Release history |
LICENSE, NOTICE |
License (FSL-1.1-Apache-2.0) |
VERSION |
Current version marker |
Everything that used to live here as file trees — the per-platform skill/command/rule
overlays and the project template — now ships inside the binary and is generated
by it: gald3r platform install <platform> writes the exact overlay for your AI tool,
always current with your binary version, on all 38 supported platforms. One binary,
every tool, no template copies to drift.
- The engine: gald3r_core (releases)
- The docs: docs.gald3r.ai
- The desktop app: gald3r_throne
- Contributing: development happens in the private engine repo — see gald3r_core for the public mirror.
FSL-1.1-Apache-2.0 — free to use; converts to Apache 2.0 on the same schedule. See NOTICE for attribution.