Who is Galaris designed for?
Galaris is designed for people working solo, project teams and small organisations building their agentic ecosystem. It helps retain agents, memory, tools and permissions while changing harnesses and model providers.
It is not designed as a large-enterprise solution for thousands of employees and tens of thousands of agents. This describes its intended audience, not a guaranteed maximum user or agent count. Discover Galaris.
Does Galaris replace my AI model?
Galaris organises work around configured models. Four text tiers divide inference duties; specialised resources handle voice and media. A harness executes a Task and is separate from the model provider.
What goes beyond a conversation?
Persistent missions, collaborating agents, revisioned documents, governed memory, recurring Goals and tracked Processes. The sourced comparison positions this combination among current AI software and harnesses.
Which tools are integrated?
The inventory lists 131 native tools in 13 families, including documents, memory, delegation, files, mail, calendars, web, Console and media. MCP connections add tools according to their own capabilities.
Is everything enabled on installation?
No. Modules, services, keys, connections and permissions determine availability. Multimedia starts active but requires compatible models; Dream learning and its four attachment analyses remain off by default. Check the models and harnesses.
Can I close my browser during a mission?
Tasks, Goals and Processes run on the server. They can continue without an open window while the instance and required services remain available. Durable waits allow resumption after a human answer or an event.
Does memory keep everything?
Recent history and durable memory have different roles. Memory has six types, access rules, validity and revisions. Permissions apply before relevance ranking. Explore memory.
Can an agent send mail on its own?
It needs a configured mailbox, permission and an explicit sending action. Task completion is insufficient. When approval is required, the complete message is frozen for the designated reviewer. Explore mail.
Do all harnesses provide the same recovery?
No. Task contracts provide a common frame, but capabilities differ. Internal effect checkpoints do not establish equivalent guarantees for every external engine. Inspect recovery.
How are costs shown?
Galaris distinguishes known, estimated, unknown and subscription-based costs. Missing data does not become zero. Budgets control phase admission; they are not an absolute cap on supplier billing.
Where can I install and try Galaris?
The installation guide on GitHub describes Docker installation from source. The Atelier Horizon kit provides a trial without external actions and a human reference. Measure benefits before claiming a saving.
Does filing a document share it?
No. Folders and icons are personal. A document retains its identity and permissions across users’ filing systems. An agent can request opening it in internal chat with document_show; read access is still required. See documents for review, filing and sharing.
Can I build a form or simulator inside a document?
Yes. An HTML document can combine text, media, CSS and JavaScript in an isolated execution environment. Text stays editable and Source opens the code. Required resources must be embedded. PDF export preserves a static copy of displayed values and results.
How do several pages share data?
They use Dataset documents containing JSON, with their own permissions and versions. Each reader needs access rights and must approve application access for the content version. Sharing a page does not automatically share the Dataset. Conflicting writes are rejected without automatic retry.
Does understanding an attachment always require a tool?
The internal engine can supply attached images, audio, video and PDFs directly to a compatible model, depending on format and transport. Dedicated tools remain useful for a saved transcript or a specific analysis; other engines do not automatically gain the same capabilities.