Multi-agent orchestration platform: 72 agents, 53 skills, 6 MCP servers, LLM provider abstraction, and 30+ data-fetching tools
Project description
Geepers
Multi-agent orchestration for LLM workflows. Ships as a Python package for building orchestrated pipelines and as a Claude Code plugin with 72 specialized agents across 15 domains.
Features
- Run hierarchical research workflows with Dream Cascade: tasks decompose into subtasks, flow through a mid-level coordinator, and arrive at an executive summary
- Dispatch parallel agents across domains with Dream Swarm: web, academic, and data searches run simultaneously and merge into a unified result
- Load config from any combination of defaults,
.env, env vars, and CLI args; later sources always win - Auto-discover API keys for 16 LLM providers from the environment with no extra setup
- Stream real-time progress via callbacks: terminal progress bars, WebSocket, or SSE
- Install as a Claude Code plugin and invoke 72 specialized agents across 15 domains from any session
Ecosystem
| PyPI | geepers-llm · geepers-kernel |
| Claude Code | /plugin add lukeslp/geepers |
| Manus | manus-skills.json in this repo |
| Codex | codex-package.json in this repo |
| Clawhub | clawhub-package.json in this repo |
| MCP servers | geepers-mcp · pip install geepers-mcp[all] |
| Orchestration | beltalowda · multi-agent-orchestration |
| Data clients | Bundled in geepers-kernel · 17+ structured APIs |
Install
As Claude Code Plugin (recommended)
In Claude Code CLI:
/plugin add lukeslp/geepers
Manual Installation (skills only)
git clone https://github.com/lukeslp/geepers.git ~/geepers
mkdir -p ~/.claude/skills
for zip in ~/geepers/skills/zips/*.zip; do
name=$(basename "$zip" .zip)
mkdir -p ~/.claude/skills/$name
unzip -o "$zip" -d ~/.claude/skills/$name
done
Quick Start
72 Specialized Agents
config = ConfigManager(app_name="myapp")
| Category | Agents | Purpose |
|---|---|---|
| Master | geepers_conductor | Routes tasks to domain specialists |
| Checkpoint | scout, repo, status, snippets, orchestrator | Session maintenance |
| Deploy | caddy, services, validator, orchestrator | Infrastructure |
| Quality | a11y, perf, api, deps, critic, security, testing, orchestrator | Code audits |
| Frontend | css, design, motion, typescript, uxpert, webperf, orchestrator | UI/UX development |
| Fullstack | db, react, orchestrator | End-to-end features |
| Hive | builder, planner, integrator, quickwin, refactor, orchestrator | Task execution |
| Research | data, links, diag, citations, fetcher, searcher, orchestrator | Data gathering |
| Datavis | color, data, math, story, viz, orchestrator | Data visualization |
| Games | game, gamedev, godot, orchestrator | Game development |
| Corpus | corpus, corpus_ux, orchestrator | Linguistics/NLP |
| Web | flask, express, orchestrator | Web applications |
| Python | pycli, orchestrator | Python projects |
| Standalone | api, scalpel, dashboard, canary, janitor, docs, git, todoist | Specialized tasks |
| System | help, onboard, diag | System utilities |
Packaged Skills
All agents are packaged as Claude Code skills in skills/zips/:
| Skill | Type | Description |
|---|---|---|
| conductor | Master | Top-level orchestrator routing to all agents |
| checkpoint | Orchestrator | Session boundaries, routine maintenance |
| deploy | Orchestrator | Deployment, infrastructure, Caddy |
| quality | Orchestrator | Code audits, a11y, perf, security, testing |
| frontend | Orchestrator | CSS, TypeScript, motion, design, UX |
| fullstack | Orchestrator | Backend + frontend end-to-end |
| hive | Orchestrator | Build from plans, execute backlogs |
| research | Orchestrator | Information gathering, API data |
| datavis | Orchestrator | D3.js, Chart.js, data visualization |
| datavis-agents | Orchestrator | Individual datavis specialist agents |
| games | Orchestrator | Game dev, Godot, gamification |
| corpus | Orchestrator | Linguistics, NLP, corpus work |
| web-dev | Orchestrator | Flask + Express web applications |
| python-dev | Orchestrator | Python projects, CLI tools |
| server-deploy | Standalone | dr.eamer.dev service management |
| engineering | Standalone | System architecture, full-stack code |
| executive | Standalone | Strategic planning, cross-team coordination |
| finance | Standalone | Monetization, financial modeling |
| product | Standalone | Product strategy, PRDs |
| api-design | Standalone | REST API design review |
| canary | Standalone | Quick health checks |
| dashboard | Standalone | System monitoring dashboard |
| docs | Standalone | Documentation generation |
| git-ops | Standalone | Git operations, conflict resolution |
| git-hygiene-guardian | Standalone | Git hygiene enforcement |
| janitor | Standalone | Project cleanup |
| scalpel | Standalone | Surgical code edits |
| todoist | Standalone | Todoist task management |
| system-diag | Standalone | Full system diagnostics |
| system-help | Standalone | Agent discovery and help |
| system-onboard | Standalone | Project onboarding |
| swarm | Research | Parallel multi-domain search |
| geepers-cascade | Research | Hierarchical deep research |
| geepers-publish | Workflow | Publishing and release workflow |
| geepers-team | Workflow | Team coordination workflow |
| data-fetch | MCP | Universal data fetching server |
| mcp-orchestration | MCP | Multi-agent orchestration server |
| vision | MCP | Alt text and image analysis |
Skill Structure
skills/
├── source/ # Skill source directories
│ ├── conductor/
│ │ └── SKILL.md
│ ├── quality/
│ │ ├── SKILL.md
│ │ └── agents/ # Sub-agent definitions
│ │ ├── geepers_a11y.md
│ │ ├── geepers_critic.md
│ │ └── ...
│ └── ...
├── zips/ # Packaged skill zips
├── ... # Individual skill directories
scripts/
├── rebuild-zips.sh # Rebuild all zips from source
└── package_all_skills.py # Generate skills from agents
Orchestrators
Agents are available via Claude Code's Task tool with subagent_type:
# Quick reconnaissance
Task with subagent_type="geepers_scout"
# Infrastructure changes
Task with subagent_type="geepers_caddy"
# End-of-session cleanup
Task with subagent_type="geepers_orchestrator_checkpoint"
# Full-stack development
Task with subagent_type="geepers_orchestrator_fullstack"
# Frontend work
Task with subagent_type="geepers_orchestrator_frontend"
Agent Categories
The bundled plugin installs a routing guide at agents/shared/AGENT_DOMAINS.md alongside the agent definitions.
Orchestrators coordinate multiple specialists:
geepers_conductor- Master routergeepers_orchestrator_frontend- CSS, React, design, motiongeepers_orchestrator_fullstack- Backend + frontendgeepers_orchestrator_hive- Build from plans and TODOsgeepers_orchestrator_quality- Audits and reviewsgeepers_orchestrator_deploy- Infrastructure changesgeepers_orchestrator_checkpoint- Session maintenancegeepers_orchestrator_research- Data gatheringgeepers_orchestrator_datavis- Visualizationgeepers_orchestrator_games- Game developmentgeepers_orchestrator_corpus- Linguistics/NLPgeepers_orchestrator_web- Web applicationsgeepers_orchestrator_python- Python projects
Related: geepers-mcp
For programmatic orchestration with LLM providers and data clients, see geepers-mcp:
- Dream Cascade - Hierarchical 3-tier research orchestration
- Dream Swarm - Multi-domain parallel search
- 12 LLM Providers - Anthropic, OpenAI, xAI, Gemini, Mistral, Cohere, etc.
- 17 Data Clients - Census, arXiv, GitHub, NASA, Wikipedia, etc.
Building Skills
# Regenerate all SKILL.md files from agent definitions
python3 scripts/package_all_skills.py
# Rebuild all zip files
bash scripts/rebuild-zips.sh
# Install to Claude Code
for zip in skills/zips/*.zip; do
name=$(basename "$zip" .zip)
mkdir -p ~/.claude/skills/$name
unzip -o "$zip" -d ~/.claude/skills/$name
done
Auto-discovers keys for 16 LLM providers from environment variables.
MCP Server Bridges
Six STDIO-based MCP servers ship with the package, covering the most common tool categories:
geepers-unified: all tools in one servergeepers-providers: LLM provider accessgeepers-data: data source clientsgeepers-cache: caching layergeepers-utility: file and text utilitiesgeepers-websearch: web search tools
Naming Registry
Maps roles to consistent identifiers across four scopes (internal, package, CLI, MCP) and resolves legacy class names to their canonical equivalents:
from geepers.naming import get_identifier, resolve_legacy
get_identifier("orchestrator", "cascade") # Returns scoped identifier
resolve_legacy("BeltalowdaOrchestrator") # Maps to canonical name
Claude Code Agents
Markdown-defined agents organized into 15 domains. Each domain has an orchestrator that coordinates its specialists; the top-level conductor routes across all domains.
| Domain | Orchestrator | Specialists |
|---|---|---|
| Master | geepers_conductor | Routes to all domains |
| Checkpoint | orchestrator_checkpoint | scout, repo, status, snippets |
| Deploy | orchestrator_deploy | caddy, services, validator |
| Quality | orchestrator_quality | a11y, perf, deps, critic, security, testing |
| Frontend | orchestrator_frontend | css, design, motion, typescript, uxpert, webperf |
| Fullstack | orchestrator_fullstack | db, react |
| Hive | orchestrator_hive | builder, planner, integrator, quickwin, refactor |
| Research | orchestrator_research | data, links, diag, citations, fetcher, searcher, doublecheck |
| Web | orchestrator_web | flask, express |
| Python | orchestrator_python | pycli |
| Games | orchestrator_games | game, gamedev, godot |
| Corpus | orchestrator_corpus | corpus, corpus_ux |
| Datavis | orchestrator_datavis | viz, color, story, math, data |
| System | (standalone) | help, onboard, diag |
| Standalone | (standalone) | api, scalpel, janitor, canary, dashboard, git, docs, todoist |
Routing: Conductor → Orchestrators → Specialists.
To invoke an agent in Claude Code, use the Task tool with the agent's subagent_type:
geepers_conductor # top-level router - when in doubt, start here
geepers_orchestrator_frontend # coordinates all frontend specialists
geepers_scout # fast project reconnaissance
geepers_orchestrator_research # coordinates research and data agents
Cross-Platform Skills
53 skills in skills/ shared across all platforms. One repo, multiple manifests. Each platform gets its own manifest file generated from the same skill directories:
| Platform | Manifest | Install |
|---|---|---|
| Claude Code | .claude-plugin/plugin.json |
/plugin add lukeslp/geepers |
| Manus | manus-skills.json |
Copy skills/ and manus-skills.json |
| Codex | codex-package.json |
Copy skills/ and codex-package.json |
| Clawhub | clawhub-package.json |
bash scripts/publish-clawhub.sh |
Regenerate all manifests after adding/removing skills:
python3 scripts/build-manifests.py
Author
Luke Steuber · lukesteuber.com · @lukesteuber.com
License
MIT - see LICENSE
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file geepers_agents-3.0.0.tar.gz.
File metadata
- Download URL: geepers_agents-3.0.0.tar.gz
- Upload date:
- Size: 10.9 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.10.12
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
31f9025f01442faf958698fd3e36f2bf0994fc7a4cc65605fc175bdf75ce8111
|
|
| MD5 |
cef7358a6392f11254c9110d63ca8ef1
|
|
| BLAKE2b-256 |
cba212a02dc492268f7eff171ab67a168d5dfac62ad6e76759ef94bcc8a793a8
|
File details
Details for the file geepers_agents-3.0.0-py3-none-any.whl.
File metadata
- Download URL: geepers_agents-3.0.0-py3-none-any.whl
- Upload date:
- Size: 7.4 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.10.12
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
af66fff9d6a04e31c301b46e20c72faf0b7f31dcecef3986d7392e6d19ac8512
|
|
| MD5 |
a95f691ef28261b3c5bc0e52dc6f778d
|
|
| BLAKE2b-256 |
b05db25a4e599688a8a11d1871eaff095b015142622058f9194b4864dd0cf4d0
|