The engine under
the magic.
gald3r_agent is the compiled gald3rCLI — file-first task/bug/idea/plan boards, thego/go-code/go-review pipeline, an agent-loop runner, and opt-in coordination with the hosted backend. One binary, every shell.
Download
Get the v0.2.0 binary
Prebuilt binaries for Windows, Linux, and macOS — each with a SHA-256 checksum and a minisign .sig signature sidecar.
Windows
x86_64 · standalone .exe
Windows
x86_64 · MSI installer
Linux
x86_64
macOS
Apple Silicon (arm64)
.sig (authenticity, key F110B9BD6FF00BA2), but the artifacts are not yet OS-code-signed — no Windows Authenticode, no Apple notarization. Windows SmartScreen and macOS Gatekeeper will therefore warn on first run (see Install below).macOS Intel: v0.2.0 does not ship an x86_64 macOS binary — install from source with the framework bootstrap (
gald3r install agent --from-source, which uses uv), or run the free framework directly — it does not require the binary.Install
Verify, then put it on your PATH
Confirm the SHA-256 checksum and the minisign .sig, then move the binary to a directory on your PATH as gald3r.
Windows (x86_64 · .exe / .msi)
# 1 — download the binary (.exe or .msi) + both sidecars$ curl.exe -LO https://github.com/Gald3r-Labs/gald3r_agent/releases/latest/download/gald3r-windows-x86_64.exe$ curl.exe -LO https://github.com/Gald3r-Labs/gald3r_agent/releases/latest/download/gald3r-windows-x86_64.exe.sha256$ curl.exe -LO https://github.com/Gald3r-Labs/gald3r_agent/releases/latest/download/gald3r-windows-x86_64.exe.sig # 2 — verify the SHA-256 checksum$ (Get-FileHash gald3r-windows-x86_64.exe -Algorithm SHA256).Hash.ToLower()$ Get-Content gald3r-windows-x86_64.exe.sha256 # the two hashes must match # 3 — verify the minisign signature (authenticity, key F110B9BD6FF00BA2)$ minisign -Vm gald3r-windows-x86_64.exe -x gald3r-windows-x86_64.exe.sig -P RWSiC/BvvbkQ8b7JpjwjDG4YUbyjBECa/t9EX/CKRe15yBuIQpV81rwA # 4 — run it (SmartScreen may warn — see the note below), or use the .msi installer$ .\gald3r-windows-x86_64.exe --versionLinux (x86_64)
# 1 — download the binary + both sidecars (.sha256 and .sig)$ curl -LO https://github.com/Gald3r-Labs/gald3r_agent/releases/latest/download/gald3r-linux-x86_64$ curl -LO https://github.com/Gald3r-Labs/gald3r_agent/releases/latest/download/gald3r-linux-x86_64.sha256$ curl -LO https://github.com/Gald3r-Labs/gald3r_agent/releases/latest/download/gald3r-linux-x86_64.sig # 2 — verify the checksum matches (must print: OK)$ sha256sum -c gald3r-linux-x86_64.sha256 # 3 — verify the minisign signature (authenticity, key F110B9BD6FF00BA2)$ minisign -Vm gald3r-linux-x86_64 -x gald3r-linux-x86_64.sig -P RWSiC/BvvbkQ8b7JpjwjDG4YUbyjBECa/t9EX/CKRe15yBuIQpV81rwA # 4 — put it on your PATH as `gald3r`$ chmod +x gald3r-linux-x86_64$ sudo mv gald3r-linux-x86_64 /usr/local/bin/gald3r$ gald3r --versionmacOS (Apple Silicon)
# 1 — download the binary + both sidecars (.sha256 and .sig)$ curl -LO https://github.com/Gald3r-Labs/gald3r_agent/releases/latest/download/gald3r-macos-arm64$ curl -LO https://github.com/Gald3r-Labs/gald3r_agent/releases/latest/download/gald3r-macos-arm64.sha256$ curl -LO https://github.com/Gald3r-Labs/gald3r_agent/releases/latest/download/gald3r-macos-arm64.sig # 2 — verify the checksum (shasum ships with macOS)$ shasum -a 256 -c gald3r-macos-arm64.sha256 # 3 — verify the minisign signature (authenticity, key F110B9BD6FF00BA2)$ minisign -Vm gald3r-macos-arm64 -x gald3r-macos-arm64.sig -P RWSiC/BvvbkQ8b7JpjwjDG4YUbyjBECa/t9EX/CKRe15yBuIQpV81rwA # 4 — not yet Apple-notarized: clear the quarantine flag so Gatekeeper allows it$ xattr -d com.apple.quarantine gald3r-macos-arm64$ chmod +x gald3r-macos-arm64$ mv gald3r-macos-arm64 /usr/local/bin/gald3r$ gald3r --versionOr: framework bootstrap
# From any gald3r-enabled project, let the framework fetch, verify the# .sha256 AND the minisign .sig, and drop the binary into the gald3r home# bin/ — no manual curl needed. Picks the right asset for your OS/arch. # preview the plan first (writes nothing)$ uv run --project .gald3r_sys/engine gald3r install agent --dry-run # apply$ uv run --project .gald3r_sys/engine gald3r install agent # ...or the IDE command wrapper (Cursor / Claude Code / etc.)> /g-install-agentThe bootstrap downloads, SHA-256-verifies, and places the binary in the gald3r home bin/. It does not auto-mutate your PATH — add that directory yourself. Flags:--release vX.Y.Z pins a version,--from-source builds from source (the Windows path),--require-verification fails closed on a checksum mismatch.
.sig for authenticity (key F110B9BD6FF00BA2), but they are not yet OS-code-signed — there is no Windows Authenticode certificate and no Apple notarization. On Windows, SmartScreen shows an “unrecognized app” prompt on first run — choose More info → Run anyway. On macOS, Gatekeeper flags it — clear the quarantine attribute (xattr -d com.apple.quarantine …) as shown above, or open it once via right-click → Open. OS signing/notarization is pending Authenticode and Apple Developer certificates.Usage
What the gald3r CLI does
A file-first task / bug / plan / coordination engine. These are the real verb groups — run gald3r --help for the full tree.
Boards — file-first & offline
gald3r status · doctor · task · bug · ideaRead and write the .gald3r/ task, bug, and idea boards straight from the shell. Deterministic, no LLM, no network — the same files your IDE agents edit.
Pipeline runners
gald3r go · go-code <id> · go-review <id>Single-shot equivalents of the @g-go pipeline: claim the next task, implement it, then run an independent verification pass — right from the CLI.
Agent loop & specs
gald3r run · agent run · swarm · autopilotRun a task through the agent loop (--task / --task-id), execute a declarative agent spec, launch a multi-teammate swarm, or drain an autopilot queue.
Coordination (hosted backend)
gald3r login · connect · claim · autoclaim · daemon · workspaceOpt-in verbs that connect to the hosted world_tree backend: authenticate, connect provider OAuth, claim tasks collision-free, and run the resident connector daemon.
Config & providers
gald3r init-providers · profile · local-model · envSeed ~/.gald3r/providers.yaml, manage your user profile, and set up keyless local models (Ollama / LM Studio) with a $0-server-cost routing entry.
Extend
gald3r plugin · skill-pack · persInstall plugins from the registry or a git URL, add and save skill packs, and switch personality packs — all from the binary.
Repo & worktrees
gald3r worktree · push-gate · housekeepCreate and merge isolated agent worktrees, run the pre-push gate, and keep the repo tidy. Absorbed from the git-commit skill family.
Desktop & traces
gald3r throne · errors · traceLaunch the installed Throne desktop app from the CLI, and inspect the agent's error/trace surface with a quick summary view.
A few real commands
# board + health — file-first, offline, deterministic$ gald3r status$ gald3r doctor # create a task, then run an implement -> review pass on it$ gald3r task add "Add rate limiting to the login route" --priority high$ gald3r go-code 42$ gald3r go-review 42 # drive the agent loop directly on a one-off task$ gald3r run --task "Fix the null check in auth.py" --provider anthropic # wire up keyless local models (Ollama / LM Studio)$ gald3r init-providers$ gald3r local-model setupCapabilities
What the Agent gives you
One binary, one command
A single compiled `gald3r` executable — no Python install, no virtualenv on the target machine. Install once, then scaffold, run, and coordinate from any shell or from inside Throne.
Deterministic where it counts
The file-first verbs — task, bug, idea, status, doctor — are deterministic and fully offline: no LLM, no network, no Docker. The agent-loop and coordination verbs (run, autoclaim, daemon) connect out only when you invoke them.
Signed & checksummed
Every published binary ships with a `.sha256` checksum and a minisign `.sig` signature (key F110B9BD6FF00BA2), verified offline by the engine's built-in Ed25519 verifier. `gald3r install agent --require-verification` fails closed on a mismatch.
Thin client, server-side loop
The shipped binary is a thin client: `gald3r run` opens a session on the hosted world_tree backend and executes server-requested local tools. Your login/token lives in the OS keyring, never in the repo.
Bring your own models
`gald3r init-providers` + `gald3r local-model setup` register locally-served Ollama / LM Studio models as keyless routing targets — $0 server cost. Cloud providers resolve keys from ~/.gald3r/providers.yaml or env.
Autonomy built in
`gald3r autopilot`, `gald3r swarm run`, and `gald3r autoclaim` drive multi-task and multi-agent work — the plumbing behind @g-go, @g-go-code, and the swarm commands, exposed as native verbs.
Download gald3r_agent v0.2.0
The precompiled binary is live for Windows (.exe / .msi), Linux, and macOS. Every asset ships with a SHA-256 checksum and a minisign .sig signature — verify before you install. The free framework does not require the binary; paid coordination features are waitlist-only while registration is closed.
The Agent binary is free to download. Paid coordination features (hosted backend, team dashboards) are waitlist-only while registration is closed.