🤖 Your AI agents. One team. Unified memory.
An open-source operating system for AI agents—bringing your agents, models, tools, and projects into one workspace.
🏠 Local-first · 🛡️ Security-focused · 🔓 Open source
Your hardware. Your accounts. Your control.
Supported provider and local-runtime families. Connection methods vary.
Mobile and smartwatch companions are on the roadmap.
Privacy and AI governance design targets—not certifications.
Formerly known as Agent Player.
🏗️ A product of Dpro. Building since May 2025.
🌐 Website · ⬇️ Download · ✅ Features · 📜 Changelog · 🤝 Contribute
Your project should not start over every time you switch assistants.
Girru brings conversations, knowledge, tasks, tools, and agent roles into a connected workspace. Organise agents into departments, give them specialties and permissions, and choose the supported models they use.
We are building Agent OS: the systems agents need to collaborate, preserve context, and carry work from your request to a result you can review.
Our direction connects everyday digital work with creative worlds, simulation, robotics, and electronics.
Talk to a local agent that coordinates your team.
Our architecture puts local coordination at the centre: understand your request, select relevant context, apply configured permissions and filtering rules, and delegate work to suitable agents.
Use supported local models alongside connected providers such as OpenAI, Anthropic, and Google. Available connection methods vary by integration and release.
You
│
▼
Local coordinator
│
▼
Permissions · Context selection · Filtering
│
┌───────────┴───────────┐
▼ ▼
Local agents Connected providers
│ │
└───────────┬───────────┘
▼
Results · Tasks · Approvals
│
▼
Controlled shared memory
Ask your team:
“Bring together the design and development departments, review this project, and propose the next steps.”
The goal is one connected workflow: discussion, findings, proposed actions, your approval, and assigned tasks.
Supported local workflows can stay on your machine. External providers receive the permitted context needed for their requests. Coordination and filtering coverage depend on the workflow and installed version.
Explain your project once. Keep its knowledge working for your team.
Memory is the core of Girru. Every agent, chat and coding tool works from the same knowledge. You don't repeat yourself, and you don't keep separate assistants up to date by hand.
- 🗂️ Organised by what it is about. You, each project, lessons learned, tasks and people each have their own place. Pinned facts apply at company, department and project level.
- 🔍 Finds the right thing. Search combines keywords with meaning and understands Arabic, including dialect. A knowledge graph links people, projects and topics.
- 🧑⚖️ Learns with your approval. Routine facts are saved on their own; important ones wait for you. Contradictions are flagged, every change is versioned, and deleted memories go to a Trash you can restore from.
- 🧾 Shows its sources. Each reply records which memories it used and why.
- 🤝 Hands work over. When a chat or task ends, a short handoff note (done, left, next) lets the next agent continue.
- 💻 One memory for your coding tools. One project brief reaches Claude Code, Codex and Antigravity together, through
AGENTS.mdand a memory server (MCP). Girru writes only inside its own marked section. - 📥 Brings in what you have. Import past coding conversations and project documents. Secrets are hidden first, and you review before anything is kept.
- 🔐 Private by design. It stays on your computer. Each agent's read and write access is set by company, department and project. Background memory work uses a local model when one is available.
- 🎯 Careful, not confident. Recall says "no reliable memory" instead of guessing, shows how sure it is, and a new fact on the same topic replaces the old one.
- 🏠 Some things never leave. Mark a memory place, such as People, to stay on this computer: only a local model ever sees it. Export everything about one person in one file, or forget it in one action; a restored backup does not bring it back.
- 🖥️ Local models only, if you like. Girru also works with no subscription: each turn fits the local model's real window. A small local memory helper, used only for search, gets ready by itself in the background once a local model runtime is installed.
Unified memory does not mean unrestricted access. Ownership, scope, and permissions remain part of the system.
See FEATURES.md for every memory feature and its status, and the storage map for what is stored where, what leaves your computer, how to export or forget it, and the numbers measured locally.
Cross-provider coverage is expanding. Sharing context does not transfer a provider’s underlying model or change its training.
Supported subscription connections require no additional API key or separate API billing.
Work through supported tools such as Claude Code and Codex, using your existing Claude or ChatGPT subscription.
API connections and local models remain additional choices.
| Connection | What you use |
|---|---|
| 🔑 Supported subscription connection | Your existing subscription, within its usage limits |
| ☁️ Optional API connection | Your own API credentials and provider billing |
| 🏠 Local model | Your hardware and a supported local runtime |
Provider support and subscription support are different: an integration does not automatically include access through a consumer subscription.
You keep your own accounts. Girru does not resell or share subscriptions. Provider terms and plan limits apply.
Connecting agents should preserve your control over information and actions.
Girru’s security approach centres on local operation, separated memory, scoped access, protected credentials, and approval gates.
| Protection | Purpose |
|---|---|
| 🧠 Memory separation | Keep private context separate and share within defined boundaries |
| 👤 Owner-scoped access | Restrict data access to the appropriate owner |
| 🔐 Agent permissions | Control access to tools and resources |
| ✅ Approval gates | Require your decision for protected actions |
| 📥 Controlled imports | Govern sources and filter sensitive information in supported import paths |
| 🔑 Credential protection | Keep secrets behind backend access controls |
| 🧹 Configured outbound filtering | Control context forwarded through supported workflows |
Local deployment gives you control over storage and connections. Cloud models and external integrations process the information sent to them.
Filtering operates within its supported scope; it does not guarantee detection of every kind of sensitive information.
Report vulnerabilities through SECURITY.md.
Girru is being developed for individuals and organisations that need control over where their AI work runs and how information is shared.
Run supported workflows on your infrastructure, manage access to knowledge and tools, and choose which external services your team uses.
Our privacy and AI governance design targets include:
GDPR / DSGVO · EU AI Act · CCPA / CPRA · UK GDPR · DPA 2018 · PECR · FADP · PIPEDA · PDPA
These are requirements we aim to support through product design and deployment controls. They are not certifications or a claim of universal compliance.
Applicable obligations depend on the organisation, jurisdiction, data, and use case. Local operation alone does not establish compliance.
A productive agent needs context, tools, responsibilities, and a way to organise its work.
Girru brings these foundations together:
| System | What it provides |
|---|---|
| 👥 Agents and departments | Teams with defined responsibilities |
| 🎯 Specialties | Configuration for particular kinds of work |
| 🎭 Personalities | Adjustable working styles and behaviour |
| 🧠 Memory | Persistent knowledge and project context |
| 💬 Conversations | Connected discussion and work |
| 📋 Tasks | Assignments, ownership, and progress |
| 🗓️ Runs and scheduling | Everything that runs in one place — logs, results and why a run stopped — plus planned and recurring workflows |
| ✅ Approvals | Owner control over protected decisions |
| 🔌 Tools and MCP | Connections to additional capabilities |
| 🧩 Apps and extensions | Specialised workflows |
| ⚙️ Model selection | Supported models chosen for different agents and tasks |
| 💻 Code workspace | Integrated project editing and agent-assisted coding; development acceptance ongoing |
| 🧑🚀 3D avatars | Compatible characters that give agents a visual identity |
| 🌍 GenVR | 3D worlds, simulation, and film production workflows in development |
| 🔊 Voice and voice cloning | Spoken interaction, voice selection, and custom voices through supported speech engines |
Availability and configuration vary by release and integration.
Build a team of distinct specialists, each with a working style suited to its responsibilities.
Configure each agent’s personality, specialty, role, department, preferred model, and permissions.
- 🎨 Personality — Adjust traits such as creativity, caution, independence, thoroughness, and communication style.
- 🎯 Specialties — Give agents focused expertise and instructions for their work.
- 👥 Roles and departments — Organise responsibilities across your team.
- ⚙️ Models and permissions — Choose supported models and control access to tools and resources.
Shared context helps your specialists collaborate while their identities and access boundaries remain distinct. Available controls depend on the installed version.
Give your team a recognisable presence alongside its personalities and specialties.
- 🎭 3D avatars — Choose an available character or upload a compatible avatar for your agent.
- 🔊 Distinct voices — Select supported voices to give different agents their own sound.
- 🎙️ Custom voices and voice cloning — Use supported speech integrations to create a personalised voice from recordings you own or have permission to use.
- 🌍 Creative continuity — Our GenVR direction connects character identity with scenes, stories, and virtual production.
Avatar formats, voice selection, and voice-cloning availability depend on the installed version, speech engine, and integration. Some speech services require separate credentials or payment; supported local engines depend on your hardware and setup.
Our goal is for your agents to build on completed tasks, your feedback, and approved lessons—preserving useful experience in memory and refining reusable skills over time.
- 🧠 Retain useful lessons — Carry relevant decisions, corrections, and task knowledge into future work.
- 💬 Learn from feedback — Use your guidance to improve future approaches.
- 🧰 Develop reusable skills — Build toward turning successful methods into repeatable instructions and workflows.
- 🔄 Share relevant experience — Make approved knowledge available to other agents within memory and permission boundaries.
- ✅ Keep you in control — Review and manage what the team retains through supported memory controls.
Memory foundations are implemented; automatic skill refinement and learning coverage are evolving. Here, learning means retaining context and improving instructions and workflows—not automatically retraining the underlying AI models.
Bring selected agents and departments into a shared discussion.
The current development implementation includes:
- 👥 Participants from multiple departments.
- 💬 Shared conversation history, attachments, and model controls.
- 🖥️ A main call or screen stage with a resizable chat sidebar.
- 🎙️ Local speech transcription with automatic pause-detected turns.
- 📷 Camera and screen previews, with image-frame attachments.
- 📋 Meeting summaries you review before proposed actions become tasks.
Ordinary chats, scheduled conversations, and meetings use the same core conversation workspace.
Real-provider voice and vision, camera hardware, and packaged platform acceptance remain under validation. Current voice interaction uses detected speech turns rather than continuous full-duplex provider audio/video streaming.
Bring conversations into your workspace through supported integrations.
| Channel | Connection |
|---|---|
| Telegram integration | |
| WhatsApp integration | |
| 🎮 Discord | Discord integration |
| #️⃣ Slack | Slack integration |
| ✉️ Email | Email integration |
| 🌐 Web chat | Browser-based conversations |
| 📞 Phone calls | Inbound and outbound calling through the Call Center extension and supported telephony providers |
Each channel has its own setup, credentials, permissions, and supported operations. Consult the documentation for your installed version.
Add capabilities through supported apps, extensions, and MCP integrations.
Specialised areas include:
- 📈 Market tools and trading workflows.
- 🔎 SEO and research.
- 🛡️ Security-related workflows.
- 🖥️ Monitoring and operational tools.
- 🌍 Creative environments and GenVR integration.
Configure what your work requires. Each integration has its own permissions, dependencies, availability, and operating requirements.
Our GenVR direction connects your AI team with interactive 3D environments.
Build scenes, arrange objects, use your own characters, and develop experiences with your agents. Control scene elements and camera composition using your mouse and keyboard.
The roadmap brings together:
- 🏗️ World building — Create environments and arrange their contents.
- 🎭 3D characters — Use your characters in stories, scenes, and interactions.
- 🎮 Interactive experiences — Develop scenarios, rules, and agent behaviour.
- 🎬 Virtual production — Stage, film, edit, and produce creative work.
- 🤖 Simulation — Build repeatable environments for testing and learning.
GenVR integration and these workflows are under development.
Work with AI to bring your characters and stories to life.
Our production vision combines storytelling, scene construction, character direction, camera work, editing, and output in one connected GenVR environment.
- ✍️ Develop ideas, scripts, episodes, and shot plans with your agents.
- 🎭 Stage scenes around your own 3D characters.
- 🖱️ Position scene elements using mouse and keyboard.
- 🎥 Direct camera angles, movement, and composition.
- ✂️ Build toward integrated editing, sound, titles, and final output.
From an idea to a scene, and from a scene to a film or series.
This describes the creative production roadmap. Individual capabilities will be documented as they are implemented and validated.
We want useful knowledge to travel with your agent—from digital work to simulated environments and supported physical devices.
Our robotics direction connects Girru’s memory with GenVR scenarios and hardware-specific integrations.
The intended workflow is to:
- 🌍 Create scenarios representing real tasks and environments.
- 🧪 Train or evaluate movement, interactions, and decisions through suitable simulation tools.
- 🧠 Preserve useful task knowledge and experience.
- 🔄 Transfer compatible memory, policies, or behaviours to supported robots.
- ✅ Validate operation on the actual hardware.
This direction includes compatible robots you already own or connect in the future, once the required integration is available.
Simulation-to-real transfer requires appropriate training methods, hardware adapters, and safety validation. Shared memory alone does not provide reliable physical skills.
Our electronics roadmap covers PCB design, Arduino, Raspberry Pi, sensors, actuators, and IoT devices.
We aim to connect the workflow:
Circuit and PCB design
↓
Firmware development
↓
Device connection
↓
Agent tools and memory
↓
Testing and controlled operation
Design the electronics, develop the software, connect the device, and integrate it with your agent’s working environment.
Integrated electronics and general robotics support are roadmap work, not capabilities guaranteed by the current release.
Girru is actively developed. Source branches and local previews can contain capabilities not yet included in published installers.
| Area | Current position |
|---|---|
| 👥 Agent workspace, conversations, tasks, memory, and configuration | Implemented; actively improved |
| 🔑 Supported subscription, API, and local-model connections | Available; setup varies |
| 🧠 Unified memory, source discovery, and automatic imports | Foundations implemented; coverage and validation expanding |
| 🤝 Meetings and automatic voice turns | Development implementation; acceptance ongoing |
| 🏠 Local coordination and outbound filtering | Coverage depends on workflow, integration, and release |
| 🌍 GenVR creative production and simulation | In development |
| 🤖 Robotics transfer and integrated electronics | Roadmap |
FEATURES.md lists every feature, section by section, with its status: built, partly built or planned.
See the changelog and release notes for the capabilities and limitations of each version.
Visit the Releases page for published installers and platform-specific notes.
For setup instructions, see the installation guide.
For source development, start with CONTRIBUTING.md.
Girru is free and open source.
Contributions are welcome across development, integrations, testing, documentation, translations, accessibility, and design.
Read CONTRIBUTING.md before opening a pull request.
💡 Have an idea or found a problem? Open an issue.
🛡️ For security concerns, follow SECURITY.md.
9MTM — Creator & Lead Developer
A product of Dpro.
MIT — see LICENSE.
Third-party components retain their respective licences. See THIRD_PARTY_LICENSES.md.