Keep more under your control.
Runtime state, local context, caches and configured credentials can stay in the environment you already manage.
Local runtime
Run Corin on your machine. Keep control close.
Corin turns your computer into the agent runtime, while every agent keeps the freedom to use the model and systems you configure.
Local runtime · Independent model choice · Your compute · Your endpoint controls
Corin workspace · this device
Runtime active
The runtime lives on your computerLOCAL COORDINATION · CONFIGURED EXTERNAL CALLS
Corin runtime
Coordination, context preparation and task state live on this device.
Business contextFiles, indexes and cached context
localTask statePending work and durable state
localAgent runtimeCoordination and orchestration
localwhen needed
Model endpointCloud or local, selected independently for each agent.
Business systemsMCP and API calls happen only when the workflow uses them.
Research agentAnalysis and synthesis
Claude
Operations agentHigh-frequency structured work
GPT
Classification agentFast routing tasks
Gemini
Document agentLong-context document work
Claude
One Corin runtime on your machine. Independent cloud-model settings for every agent.
Your machineCorin runtime
More stays inside the environment you manage.
Local context, task state, caches and runtime processing stay with the machine until a configured workflow needs an external model or system.
Your endpoint policy and network controls remain part of the boundary.Why local-first
Keep the runtime close to where work happens.
Corin can work from a machine you already control instead of adding another hosted runtime between your agents, your data and the systems they use.
Use the hardware you already own.
Retrieval, indexing, caching, context preparation and orchestration can use the machine you already own.
Move only what needs to move.
Prepare context locally, then call the model or business system the workflow actually needs.
Skip unnecessary round trips.
Local files, caches and compute are available directly to the runtime.
Security and control
Keep more of the operating boundary in your environment.
Corin runs from a machine you manage. You decide which models and business systems each agent can reach, while local context and task state stay close to the runtime.
1
Control the data path.Local processing stays local; external calls happen through the connections you configure.
2
Keep credentials with the runtime.Model keys and system access stay under your endpoint and network controls.
3
Reduce central copies.The runtime does not require Corin to hold another hosted copy of every working context or task state.
Your machine · controlled boundary
Model endpointOnly the configured model call crosses the boundary.
Business systemOnly the workflow's permitted API or MCP call leaves the runtime.
The Corin runtime stays on the machine you manage; configured calls leave only when the workflow needs them.
Model routing · same Corin runtimeRESEARCH AGENT · CLAUDE
R
Research agentAnalysis and synthesis
O
Operations agentStructured operational work
C
Classification agentFast routing and triage
Assign selected agent:
Model freedom
Local runtime. Independent model choice.
Where Corin runs and which LLM an agent uses are separate decisions. Corin runs on your machine while each agent can use the cloud model best suited to its role.
Model selection depends on your edition. Starter uses one Corin-selected model; Community, Pro and Enterprise support per-agent selection. Compare editions.
Deployment does not lock the model. The app stays local while Corin routes each agent to the model endpoint and credentials you configure.
Use your own compute
Your machine becomes part of the stack.
Corin uses your machine for retrieval, indexing, caching, context preparation and orchestration. LLM calls still go to the model providers configured for each agent.
Work handled on your machine
RUNTIME VIEW · ILLUSTRATIVEAgent orchestration
device
Task state
device
Retrieval
device
Caching
device
Runtime: core coordination work uses the machine you already have.Model inference remains with the provider selected for each agent.
Long-running work
Keep work moving on the machine that fits the job.
Use a workday computer for interactive agents, or an always-on machine for scheduled and longer-running work. Durable task state preserves where the workflow is when execution pauses.
Task starts
Corin gathers local context and begins the workflow.
Work waits
The task can remain open while a person or external event is pending.
Signal arrives
The same work item continues from the pending step.
Next check runs
An available machine carries the workflow into the next scheduled action.
Durable state keeps the workflow ready for its next step.While work is waiting on a person, a schedule or an external event, Corin keeps the saved step and resumes from there.
Before you install
System requirements
Run the Corin app on a supported version of macOS, Windows or Linux.
macOS
macOS 13 (Ventura) or higher
Windows
Windows 10 or higher
Linux
Ubuntu 22.04 LTS+ or Debian 12 (bookworm)+, on x64 or arm64
Run Corin where your work already happens.
Keep the runtime on a machine you control, connect the models and systems you choose, and let your agents work from there.