Connect local AI apps
Stellary can use the AI CLIs and local servers installed on your machine to run agent missions from the workspace. Stellary keeps the work coordinated in the product while execution happens close to your local repository.
Mode
Local execution
Target
Compatible or dedicated machine
Context
Workspace, card, and linked repo
Overview
The local runtime connects the workspace, the desktop app, and the installed AI runner. Stellary keeps missions, approvals, and history; the local machine runs the authorized model.
Workspace mission
The card, agent instructions, and useful context stay in Stellary.
Desktop host
Stellary announces ready models, linked repositories, and machine state.
Local runner
The CLI or local server handles the mission with the selected model.
Product trace
Progress, errors, interruptions, and results come back into Activity.
Set up Stellary
Configuration happens on the machine that owns the AI CLIs, local servers, and working repositories.
- 01
Open Stellary
Sign in with the same account you use in Stellary Cloud, then select the target workspace.
- 02
Connect AI apps
Detect or connect Codex, Claude Code, Cursor Agent, LM Studio, AnythingLLM, or another available runner.
- 03
Share models
Share only the models workspace agents may use. Other models stay private.
- 04
Link the local repository
Add the local Git folder and attach it to the matching Stellary project.
- 05
Refresh
Refresh detection after a connection, CLI login, or repository change.
Supported local apps
The catalog is organized by provider. It only shows models that are ready and usable for the current workspace.
Agent CLIs
These runners start an installed local tool and can work inside the repository when the project is linked.
Local AI servers
These apps expose a local API. They are useful for local responses; some do not edit the repository.
Workspace catalog
Stellary builds the catalog from connected machines. A model is selectable only when it is ready, visible to the user, and authorized for the workspace.
Controlled visibility
You see your own models. Other members only see models shared with the workspace.
Machine state
An offline, paused, or disconnected machine cannot receive a mission.
Model options
When the runner supports it, Stellary exposes the available reasoning level.
Configure an agent
Local mode is configured in the agent settings. It replaces the Cloud runtime for that agent’s missions without changing rules, tools, autonomy, or approvals.
- 01
Choose the execution mode
Select Local execution. The local model panel appears below the field.
- 02
Choose the target
Compatible machine lets Stellary pick an available host. Specific machine forces one computer.
- 03
Choose the model
Unavailable models show a reason: offline, missing repository, connection required, or setup required.
- 04
Set reasoning
If the model supports it, choose the reasoning level that fits the work.
Mission execution
When a card starts a mission for a local agent, Stellary keeps the mission in the workspace and the desktop app handles it as soon as an eligible machine is available.
- 01
Mission created
The card, project, agent instructions, and context are prepared in Stellary.
- 02
Host checked
Stellary checks the model, machine state, workspace access, and linked repository.
- 03
Runner started
Stellary starts the CLI or calls the local server with mission context.
- 04
Result returned
Summary, changed files, interruptions, and readable errors come back into Stellary.
Control & safety
Local runners add power, so Stellary keeps explicit controls around access, visibility, and execution.
Same autonomy
A local agent still follows its autonomy mode: autonomous, supervised, or approval-based.
Voluntary sharing
A local model is not offered to other members until it is shared with the workspace.
Repository required
Runners that edit code must be linked to the project.
Troubleshooting
The states shown in the interface match the checks used before execution.
No local model detected
Open Stellary, connect the local runner, check that the CLI or server responds, then refresh.
Model not visible in the agent
Share the model from the Models tab in the Local runners panel, then sync.
Machine offline or paused
Keep Stellary open, select the right workspace, and enable local execution.
Local repository missing
Add the repository in the Repositories tab and link it to the Stellary project.
Connection or setup required
Start the runner login, or configure the local URL/API key for apps that need one.
Related guides
These pages complete the local runtime story: agents, MCP, getting started, and desktop installation.