Song magic for your codebase.
Persistent memory, multi-repo orchestration, and adversarial quality gates for AI coding agents — across every major IDE. Without changing how you code.
v3.0.0 · 38 AI tools · FSL-1.1-Apache · file-first, optional engine
- 13
- Agents
- 111
- Skills
- 179
- Commands
- 38
- Platforms
Works with your IDE
The Problem
Your AI agents forget everything the moment you close the chat.
Session 1: You explain your architecture, your constraints, why you chose Postgres over Mongo.
Session 12: The agent uses SQLite. Because it doesn't remember session 1.
You're running 4 repos — api, web, mobile, shared-lib. When something changes in shared-lib, every downstream repo needs to know. There's no mechanism for that.
You ask the same agent that wrote the code to verify it. It passes everything — of course it does. The bugs it introduced are invisible to it.
Gald3r is the infrastructure layer that fixes all three. It wraps around your AI IDEs and gives agents durable memory, cross-repo coordination, and adversarial quality enforcement.
What Gald3r Gives Your Agents
Four pillars of AI-assisted development
The gald3r family
Three products. One ecosystem.
The file-first framework is free and ships today. The runtime and desktop app are downloadable now — paid coordination features are waitlist-only.
gald3r
The free, file-first framework — 13 agents, 111 skills, 179 commands, rules, hooks, and a Python engine. Plain markdown in .gald3r/, works in 38 IDEs.
gald3r_agent
The Python runtime that powers the gald3r CLI. Tool loop, inference backends, tracing/eval, MCP routing. Ships as a signed packaged binary — no source build required.
gald3r_throne
Hlidskjalf — the desktop workspace control plane. A Tauri 2 app with live task dashboards, dry-run→apply with backup/rollback, personality packs, animated avatars, and a Chrono scheduler.
Download v0.2.0 →New in v2.0.0
The Engine — a deterministic core, no LLM required
Gald3r v2.0.0 shipped a bundled Python engine that handles all the mechanical parts of task management without touching your AI or your network. No Docker, no servers, no API keys until you want the optional self-hosted MCP backend.
gald3r initScaffold or upgrade a project in one command. Idempotent — safe to re-run.gald3r task newCreate structured task files with acceptance criteria, type, and priority.gald3r doctorRead-only health check — 11 gates, suitable for CI pipelines.gald3r upgradePull the latest release; takes a timestamped backup first; auto-rolls back on failure.gald3r vault ingestIndex local files into the knowledge vault without Docker.
$ gald3r init
✓ .gald3r/ structure created
✓ TASKS.md, BUGS.md, PLAN.md ready
✓ 13 agents, 111 skills deployed
$ gald3r doctor
Running 11 health checks...
✓ All checks passed
$ gald3r upgrade
Backup → .gald3r/backups/20260620_133000
Applying v3.0.0 patches...
✓ Upgrade complete. Tasks and bugs untouched.
New in v2.0.1
Upgrade without fear
gald3r upgrade is designed so the worst-case outcome is a 30-second rollback. Your tasks, bugs, plans, and vault are never touched.
Timestamped backup
Before touching a single file, the engine snapshots your .gald3r/ directory to .gald3r/backups/YYYYMMDD_HHMMSS/.
Selective patch
Only framework files are updated — agents, skills, commands, hooks. TASKS.md, BUGS.md, PLAN.md, and vault notes are untouched.
Auto rollback on failure
If any patch step fails, the engine automatically restores the backup. No partial upgrades, no corrupt state.
How far we've come
18 releases. 6 months. Still accelerating.
Changelog
Actively maintained and evolving
Latest releases across the gald3r family — the framework, the Agent CLI, and the Throne desktop app. Scroll each column to see full version history.
Release History
- ›One-command install: `gald3r install agent|throne|all` (Windows, macOS, Linux)
- ›Centralized install home (settings, logs, vault) + global `gald3r` command from any directory; USB-portable mode
- ›Safe self-update: `gald3r upgrade` with timestamped backup + automatic rollback; tasks/bugs/plans never touched
- ›`gald3r init` / `gald3r update` — idempotent project scaffolding & upgrade
- ›Canonical hook event set (6 events, one shared Python core) wired across 9/9 hook-capable platforms
- ›Build hygiene: generated repos no longer ship .venv / __pycache__ / .pytest_cache
- ›Self-update stack: `gald3r version-check` + `gald3r upgrade` (offline-first, rollback-safe)
- ›gald3r_throne in-app update — detect & apply updates inside the app, in compiled Rust (no Python)
- ›Throne developer-experience suite
- ›The Engine Release — bundled file-first Python engine (.gald3r_sys/engine/): deterministic, no LLM, no network, no Docker
- ›`gald3r` CLI + MCP server (~37 tools): `gald3r task new`, `gald3r bug new`, `gald3r vault ingest`, and more
- ›`gald3r doctor` — read-only health check + CI gate
- ›Five maintenance scripts absorbed into engine verbs; each keeps a no-engine .ps1 fallback
- ›Judgment/prompt layer: 15 reasoning assets served by the engine
- ›Plugin system foundation: @g-plugin-install (ADR-015), manifest schema + validator, security-first installer
- ›gald3r public repo restructured — 34 platform dirs moved into platforms/ subfolder
- ›post_push_verify.ps1 — 11-check post-release gate (incl. secrets scan)
- ›Removed redundant @g-kamikaze / @g-juggernaut aliases (use @g-mission)
- ›35 AI tool support finalized — mimo-code (Xiaomi MiMo-Code, OpenCode fork) added as supported platform
- ›Platform CRASH capability matrix published across all 35 tools
- ›Full platform spec HTML generation pipeline
- ›34 AI coding tools — 11 new platforms (Kimi Code, TRAE, Qoder, Amp, Continue, Void, and more), all QA-passing
- ›g-go-go coding swarm fix — Phase 1 now runs N parallel coders via g-go-code-swarm
- ›Context-aware throttle ON by default — reduces parallelism under context pressure instead of stopping
- ›Platform templates restructured to repo root as drop-in folders (T1522)
- ›Schema enforcement — every .gald3r/ file validated against 15 versioned schemas; sub-second session-start drift probe
- ›Semantic versioning + release management — g-skl-ship / @g-ship promote CHANGELOG to tagged releases
- ›Dedicated bug-fix pipeline (@g-go-bugs)
- ›WPAC v1.6 autonomous-child workspace model; encoding normalization
- ›HTML / JSON / TOON output modes + 3-theme system (--html / --json / --toon)
- ›Memory compression + project vocabulary management
- ›Skill review (@g-skill-review) + opt-in auto-skill-generation after PASS verdicts
- ›Plugin author guide (docs/PLUGINS.md) + skill-pack distribution and naming guidelines
- ›Platform Framework Architecture (.gald3r_sys/) — canonical framework tree
- ›g-skl-security-scan — two-phase SAST scanner across 12 vulnerability categories
- ›g-skl-context-builder + g-skl-delegate workflow skills
- ›Recon & research suite expansion
- ›GitHub Copilot support — 6th IDE, 89 commands deployed to .copilot/
- ›Recon suite: g-skl-recon-repo/url/docs/yt/file replaces ingest/harvest/reverse-spec
- ›Research workflow: g-skl-res-review → res-deep → res-apply (3-step pipeline)
- ›Swarm commands: @g-go-swarm, @g-go-code-swarm, @g-go-review-swarm
- ›Release management: g-skl-release + .gald3r/releases/ + RELEASES.md
- ›g-skl-medic: surgical repair for individual .gald3r/ files
- ›Codex + Copilot CLI skills added
- ›Added g-skl-pcac-send-to for topology-aware file transfer
- ›g-skl-medkit now handles v1.0→v1.1→v1.2 migration in one pass
- ›Fixed TASKS.md sync drift in multi-agent sessions
- ›Added g-skl-cli-gemini checkpoint flag documentation
- ›Feature pipeline: staging→specced→committed→shipped lifecycle
- ›g-skl-reverse-spec: 5-pass deep repo analysis (now g-skl-recon-repo in v1.4)
- ›OpenCode IDE support added (5th IDE target)
- ›12 pre-commit/push hooks including PCAC inbox check
- ›PCAC multi-project coordination system (10 skills)
- ›Parent/child/sibling topology with bidirectional linking
- ›Task broadcast cascade with configurable depth (1-3)
- ›INBOX.md cross-project message routing
- ›Initial release — 9 agents, 47 skills, 76 commands
- ›5-IDE parity: Cursor, Claude Code, Gemini, Codex
- ›Adversarial two-phase code review gate
- ›File-first architecture — zero external dependencies
Release History
- ›One-command install: `gald3r install agent|throne|all` (Windows, macOS, Linux)
- ›Centralized install home (settings, logs, vault) + global `gald3r` command from any directory; USB-portable mode
- ›Safe self-update: `gald3r upgrade` with timestamped backup + automatic rollback; tasks/bugs/plans never touched
- ›`gald3r init` / `gald3r update` — idempotent project scaffolding & upgrade
- ›Canonical hook event set (6 events, one shared Python core) wired across 9/9 hook-capable platforms
- ›Build hygiene: generated repos no longer ship .venv / __pycache__ / .pytest_cache
- ›Self-update stack: `gald3r version-check` + `gald3r upgrade` (offline-first, rollback-safe)
- ›gald3r_throne in-app update — detect & apply updates inside the app, in compiled Rust (no Python)
- ›Throne developer-experience suite
- ›The Engine Release — bundled file-first Python engine (.gald3r_sys/engine/): deterministic, no LLM, no network, no Docker
- ›`gald3r` CLI + MCP server (~37 tools): `gald3r task new`, `gald3r bug new`, `gald3r vault ingest`, and more
- ›`gald3r doctor` — read-only health check + CI gate
- ›Five maintenance scripts absorbed into engine verbs; each keeps a no-engine .ps1 fallback
- ›Judgment/prompt layer: 15 reasoning assets served by the engine
- ›Plugin system foundation: @g-plugin-install (ADR-015), manifest schema + validator, security-first installer
- ›gald3r public repo restructured — 34 platform dirs moved into platforms/ subfolder
- ›post_push_verify.ps1 — 11-check post-release gate (incl. secrets scan)
- ›Removed redundant @g-kamikaze / @g-juggernaut aliases (use @g-mission)
- ›35 AI tool support finalized — mimo-code (Xiaomi MiMo-Code, OpenCode fork) added as supported platform
- ›Platform CRASH capability matrix published across all 35 tools
- ›Full platform spec HTML generation pipeline
- ›34 AI coding tools — 11 new platforms (Kimi Code, TRAE, Qoder, Amp, Continue, Void, and more), all QA-passing
- ›g-go-go coding swarm fix — Phase 1 now runs N parallel coders via g-go-code-swarm
- ›Context-aware throttle ON by default — reduces parallelism under context pressure instead of stopping
- ›Platform templates restructured to repo root as drop-in folders (T1522)
- ›Schema enforcement — every .gald3r/ file validated against 15 versioned schemas; sub-second session-start drift probe
- ›Semantic versioning + release management — g-skl-ship / @g-ship promote CHANGELOG to tagged releases
- ›Dedicated bug-fix pipeline (@g-go-bugs)
- ›WPAC v1.6 autonomous-child workspace model; encoding normalization
- ›HTML / JSON / TOON output modes + 3-theme system (--html / --json / --toon)
- ›Memory compression + project vocabulary management
- ›Skill review (@g-skill-review) + opt-in auto-skill-generation after PASS verdicts
- ›Plugin author guide (docs/PLUGINS.md) + skill-pack distribution and naming guidelines
- ›Platform Framework Architecture (.gald3r_sys/) — canonical framework tree
- ›g-skl-security-scan — two-phase SAST scanner across 12 vulnerability categories
- ›g-skl-context-builder + g-skl-delegate workflow skills
- ›Recon & research suite expansion
- ›GitHub Copilot support — 6th IDE, 89 commands deployed to .copilot/
- ›Recon suite: g-skl-recon-repo/url/docs/yt/file replaces ingest/harvest/reverse-spec
- ›Research workflow: g-skl-res-review → res-deep → res-apply (3-step pipeline)
- ›Swarm commands: @g-go-swarm, @g-go-code-swarm, @g-go-review-swarm
- ›Release management: g-skl-release + .gald3r/releases/ + RELEASES.md
- ›g-skl-medic: surgical repair for individual .gald3r/ files
- ›Codex + Copilot CLI skills added
- ›Added g-skl-pcac-send-to for topology-aware file transfer
- ›g-skl-medkit now handles v1.0→v1.1→v1.2 migration in one pass
- ›Fixed TASKS.md sync drift in multi-agent sessions
- ›Added g-skl-cli-gemini checkpoint flag documentation
- ›Feature pipeline: staging→specced→committed→shipped lifecycle
- ›g-skl-reverse-spec: 5-pass deep repo analysis (now g-skl-recon-repo in v1.4)
- ›OpenCode IDE support added (5th IDE target)
- ›12 pre-commit/push hooks including PCAC inbox check
- ›PCAC multi-project coordination system (10 skills)
- ›Parent/child/sibling topology with bidirectional linking
- ›Task broadcast cascade with configurable depth (1-3)
- ›INBOX.md cross-project message routing
- ›Initial release — 9 agents, 47 skills, 76 commands
- ›5-IDE parity: Cursor, Claude Code, Gemini, Codex
- ›Adversarial two-phase code review gate
- ›File-first architecture — zero external dependencies
FAQ
Questions developers ask
- What is gald3r?
- Gald3r is an AI-native task management framework that wraps around your AI coding IDE. It gives AI agents persistent memory, multi-repo orchestration, and adversarial quality gates — so they remember your architecture across sessions, coordinate changes across repositories, and cannot sign off on their own work.
- How do I install gald3r?
- Clone the repository and copy the template into your project: git clone https://github.com/Gald3r-Labs/gald3r.git. Gald3r is file-first — no servers or subscriptions are required for the free framework. The Quickstart walks through your first task in under 10 minutes.
- Which IDEs does gald3r support?
- 38 AI coding tools are supported. Six have full parity — Cursor, Claude Code, Gemini CLI, OpenAI Codex, OpenCode, and GitHub Copilot — sharing identical agents, skills, commands, and task state, so you can switch between them mid-task and pick up exactly where you left off.
- Is gald3r free?
- Yes. The complete gald3r framework — all agents, skills, commands, rules, hooks, the vault, and multi-project coordination — is free under the FSL-1.1-Apache license (Functional Source License, which converts to Apache-2.0 after two years). There are no tiers. An optional self-hosted Docker backend adds MCP memory/search tools and is also free.
ᚷᚨᛚᛞᚱ
Ready to forge your workflow?
Gald3r public releases use the FSL-1.1-Apache license posture. Install it in any project in minutes — no external services, no subscriptions, no magic until you copy the files.