A task to carry out
Research, analyse, write: the harness drives the agent’s work as it adapts its steps and uses authorised tools to reach the goal.
Explore agentic tasksThe hub for your agentic ecosystem
Your agentic ecosystem, self-hosted.
Change the engines. Keep your ecosystem.
Ideal for solo work, project teams and small organisations.
Open source · CeCILL 2.1Talk to your agents like colleagues. They can handle tasks or start your predefined business processes. Run your models entirely locally, in the cloud, or combine both. Memory, tools and permissions stay together in Galaris, through its applications or your own.
Fully local, cloud or hybrid inference: your choice.
From conversation to action
Talk to your agents like colleagues.
Galaris has two agentic loops. The conversation loop uses history and memory, with a limited set of tools to stay focused on the exchange. The task loop executes work through the harness; business processes handle predefined workflows. You keep talking to your agent while it works, without mixing the discussion with execution details.
A business process to start
The agent starts a configured business process assigned to it. Steps, rules and approvals are defined in advance; Galaris tracks the execution.
Explore business processes with n8nKeep the conversation flowing while work continues. “How’s it going?” The agent checks the actual progress and answers naturally without interrupting its task. You can also start a task, request a stop or adjust its instructions along the way, depending on its state. You customise instructions, models, harnesses and the tools available to your agents. Understand the two loops
Need a model inside the workflow? The process can call the Galaris inference API, with call tracking and available cost data. Explore the API integration
The whole platform
Choose the work to delegate.
Six starting points to find the capabilities your team needs, their screens and their requirements.
Communicate
Reach agents through your conversations.
Delegate and follow
From individual tasks to lasting goals.
Produce
Deliverables to review, edit and share.
Organise
Build on work already completed.
Connect
Choose models, engines and services.
Administer and evaluate
Inspect access, results and costs.
From document to working tool
Build and use your tools, inside chat.
A form to complete, a simulator to adjust, a 3D scene to explore: open the document beside the conversation and keep working with your agent.
- Enter information in a form connected to a Dataset.
- Adjust a mini-application’s inputs and see the result.
- Manipulate a 3D view embedded in the document.
What your agents can do
One place to talk, act and keep work moving.
From the first message to background work, explore the features that make a difference.
01Missions & plans
Keep talking. Your missions keep moving.
Give an agent a lasting piece of work: the original request and attachments stay separate from supplementary context. Inspect its plan, steps and result while the conversation continues.
Use it toPrepare a document in several stages and return to check its progress.
Explore this feature
02Agent collaboration
Bring specialists together on the same job.
An agent can delegate a contribution to another, wait for its response and resume work with the result. Contributions remain linked to the mission.
Use it toHave one agent research sources and another prepare the summary.
Explore this feature
03Graph memory
See how your information connects.
Find a passage in your documents, then explore links to sources, attachments and topics. The graph preserves identities and access rights; acquired descriptions complement readable content.
Use it toReturn to a project and find the documents and discussions behind it.
Explore this feature
04Dream mode
Build on work already done.
Dream maintains and consolidates memory in the background. You can inspect its operations; skill learning offers another way to build on experience when enabled in the configuration.
Use it toOrganise accumulated context and inspect what has been consolidated.
Dream skill learning is disabled by default.
Explore this feature
05Voice conversations
Talk to your agents.
Speak with your agents using a compatible model. Session monitoring lets you inspect voice activity and the tools used.
Use it toDiscuss a request out loud.
Explore this feature
06Goals & cycles
Stay on track over time.
Define a goal to pursue over time. Cycles retain their work, evaluations and evidence, with a configurable schedule.
Use it toFollow a research topic and its new questions in each cycle.
Explore this feature
07Business processes
Start workflows from the conversation.
Your agent can start an authorised process configured on your instance. The request stays linked to the business workflow and its progress.
Use it toStart the processing workflow for a document.
Explore this feature
08Agents, models & skills
Build the team that suits you.
Assign roles, model profiles, skills and tools to your agents. Choose the capabilities each agent needs and connections to your services.
Use it toEquip a research agent and a writing agent differently.
Explore this feature
What you build with Galaris
Your engines change. Your work stays.
Agent identities, memory, connections and permissions stay in Galaris. Evolve your ecosystem’s engines while keeping what you have built.
Change the harness or model. Keep the rest.
Choose fully local inference, cloud providers or both. Change compatible models or harnesses: your agents, their memory, connections and permissions stay managed in Galaris.
Your research agent keeps its role, knowledge and authorised tools when you choose another harness or model provider.
Build on useful knowledge from your missions.
Documents, sources and conversations feed connected memory governed by access rights. You and your agents work with the same document from drafting through review and sharing.
Return to a project and its sources to prepare your next decision.
Give your team a goal that outlasts the chat.
Set a direction and a schedule. Work cycles keep their results, evidence and evaluations to determine what comes next. A designated contact can be asked for a decision when needed.
Follow a research topic and review progress from one cycle to the next.
Delegate while staying in control.
Follow your agents’ steps, results and actions. When delivery can be retried, the existing output is reused. Recovery takes account of known effects and uncertain situations.
Retry a supported delivery without asking the agent to recreate the document.
Build on your agents’ experience.
Dream maintains memory and can derive skills from past missions.
Find useful knowledge and review skills proposed from completed work.
Skill learning is disabled by default. You choose whether to enable it and evaluate its results.
A shared system across the layers
Build your agentic ecosystem.
Galaris is the hub that preserves your working environment as its engines evolve. It connects models, harnesses and tools, and also provides its own chat, internal harness and applications.
Galaris: connections, permissions and tracking
Configure and authorise your tools in Galaris. Connect an MCP server, define which functions each agent can access, then choose its execution engine. Galaris presents the authorised catalogue to that engine. Tool configuration stays in the hub.
- Applications
Conversations, goals, documents and processes: the spaces where work takes shape.
- Shared services
Memory, skills and MCP configured in Galaris: functions exposed according to each agent’s permissions.
- Agent execution
Agents, roles and missions stay in Galaris; choose an execution harness suited to the work.
- Model access
API gateway, inference tracking, events, usage and available costs.
On the infrastructure and models you choose
- Models and providers
- Fully local inference on your servers, cloud providers or a mix of both, using compatible models and protocols.
- Infrastructure
- Servers, compute, storage and networking for your deployment.
Reach your agents where you already work.
In Galaris applications
Chat, follow your goals and work on documents in the shared workspace.
In your messaging channels
Nextcloud Talk, Matrix, OneBot, Telegram: talk to your agents through configured bridges, with each channel’s supported capabilities.
In your clients and applications
Configure Janus as a single model in a compatible client. Select an agent with @code and Janus hands over. APIs also provide access to models and agents.
For example, your business application can consume a model through Galaris and retrieve the associated inference without using Galaris chat.
Stay in control
Autonomy you can inspect.
Choose your agents’ access, follow useful steps and review the result before using it.
Understand the controls
- Tools assigned to each role
- Inspectable states and results
- Sources and limitations to check
Your first test case
A concrete starting point. A result to compare.
Atelier Horizon is preparing a mobile repair service. Three synthetic notes, a contradiction and a decision give you a focused way to evaluate Galaris.
- 01
Ask
Request a launch brief with sources and open questions.
- 02
Check
Does the agent identify contradictory information?
- 03
Decide
Provide the correction and compare the result with the reference.
Download the test kitMarkdown · no registration
The kit contains a brief, sources and a human-written reference. It is not a result already produced by Galaris.
Start at your own scale
One agent. One model. Your first mission.
Find installation instructions on GitHub, then prepare your first trial with the Atelier Horizon kit.
Your team, your knowledge, your goals. Explore Galaris’s strengths