Work computers
The Guild interface, agents and project files. The model does not need to run on every laptop
Guild Enterprise & On-PremCorporate pilot
Your AI team.Inside your network
Agents, working data and the model itself run in your company's infrastructure. We will develop a fully local Guild configuration without calls to OpenAI, Anthropic or other cloud AI APIs
We assemble and validate the isolated configuration during a pilot
Your data
Projects and context inside your company network
Your model
Answers generated on your own hardware
Your rules
Access and connections under your control
01 / Inside your network
The entire task stays in your network
An employee manages agents in Guild. The agent works on the project, an internal gateway forwards requests to the model and your company's server generates the response. We validate this architecture in the pilot
Your company infrastructureTarget pilot architecture
Computer running Guild
Agent team, files and task execution
WorkstationRequest and response
Internal gateway
One connection point between agents and the model
Compatible APILocal network
Model server
DeepSeek or another model selected for the pilot
Your hardwareGuild services stay inside, too
- Portal and relay
- Activation and licensing
- Updates
- Internal integrations
Company servers
The model, API gateway and Guild services. We size capacity for your context and concurrent agent load
Internal tools
Your repositories, documents and MCP services. Connections depend on the tasks and network policies
02 / The team at work
All agent work in view
Tasks, plans, decisions and results share one workspace. We bring this management approach to a corporate configuration with a local model
Distribute the work
Multiple agents. One result
Assign parts of a project to agents and follow their progress. The pilot will establish how many parallel tasks your model and server can handle
Stay in control
Important decisions go through you
Review questions, approve actions and return to results. Permission handling and session continuation are part of the pilot test
03 / Isolation testing
Internet disconnected. Work continues
This is an acceptance criterion for the isolated pilot. We test the complete workflow with external access blocked, including startup, tools and session recovery
What we test
- The model responds from your server
- The portal, relay and device access work inside the network
- Activation and licences are checked internally
- Updates use the agreed delivery method
- Integrations and service traffic need no external access
We validate results through completed tasks and network checks. The final delivery scope and limitations are documented
04 / Security and administration
The company sets the rules. Employees cannot switch them off
Data, device, access and spend controls are part of the enterprise configuration of Guild. Whatever your security team checks goes into the pilot acceptance criteria
Secrets and personal data never reach the model
Keys, passwords, personal data and labelled documents are intercepted before they are sent: warn or block, by company policy. Every incident is logged — who, which rule, what was found, which model
Tools, models and network by allow-list
Administrators define allowed tools, models, MCP servers, domains and agent permission modes. Employees cannot lift these limits; agent commands run inside a file and network sandbox
Every device is known and revocable
Macs and iPhones are enrolled in the company cabinet and revoked with one click; connections use mTLS. Sign-in through corporate SSO, seats and roles through SCIM and directory groups
Who, what, when and from which device
Agent actions, human decisions and policy changes go to the log, with SIEM export and retention on your terms. Conversations with agents stay on the employee's computer: the company server sees events and metadata
Effectiveness without surveillance
Who uses it, how many tasks and sessions, accepted changes and time saved, per team. Individual figures open by role and are themselves logged; prompts never enter reports. That keeps legal review straightforward under labour law and GDPR
Seats, limits and budgets, centrally
Accounts are corporate: the provider's no-training, no-retention terms cover the whole organisation, and personal subscriptions stay outside the perimeter. Administrators grant and revoke seats, limits apply per team and per person, token and money spend is visible per model, and overspend triggers an alert
For Russian customers
Regulatory requirements are met by the delivery scope, not by promises
Guild does not yet hold SOC 2 or ISO 27001 certification. Compliance is demonstrated by the delivery scope, the audit log and pilot results on your infrastructure
- Federal Law 152-FZ: personal data is processed in Russia, with the panel and relay inside your perimeter and a local or Russian model
- FSTEC Order 117: restricted information is never passed to the model vendor; requests and responses are kept in the log for at least a year
- Request and response filtering, MFA and a dedicated network segment are part of the isolated delivery
- For banks: blocking DLP and a model register per GOST R 57580 and Bank of Russia guidance
05 / From task to installation
Start with your real workflow
We select a local model and configuration for your work, then complete the full cycle: from the first agent's task to testing the team inside an isolated network
- 01
Start with your task
Choose a real project, participants and a clear result: a code change, document review or prepared materials
- 02
Build the configuration
Select the model, server, agent engine and internal tools. Agree on delivery, access and updates
- 03
Test it at work
The agent reads files, makes changes, runs checks, requests permission and resumes its session. We measure team load
- 04
Disconnect the internet
Repeat the agreed workflow with external access blocked. Check network traffic, startup and recovery
The outcome: a validated configuration
Task and isolation test results, load measurements, server requirements and agreed terms for further deployment
06 / Delivery details
Let's discuss your infrastructure
Local models, isolated networks, devices and pilot terms
Guild is a management layer and unified interface for a team of AI agents. It helps distribute tasks, shows progress, brings decisions together and lets you manage the process from Mac and iPhone. It is useful for leaders, product teams and functional teams that have outgrown individual terminals
Enterprise is a corporate Guild configuration tailored to your company's tasks and policies. Models can run locally or connect through an agreed API. The primary on-prem scenario puts the model and Guild services inside your infrastructure, with operation tested without external access. We define the components and connections for each pilot
Yes, we propose exploring this in a corporate pilot: Guild, a local model, a gateway and supporting services inside your company network. Offline operation is a pilot acceptance criterion. A complete isolated deployment still needs to be assembled and verified in your infrastructure
The isolated configuration aims to operate without calls to those cloud APIs. A model on your server should generate the answers. External integrations and service traffic are disabled or replaced by internal alternatives. We verify this through traffic inspection and operation with internet access blocked
DeepSeek is a candidate for the pilot. We select and test the model version, inference server and compatible agent engine together. We check task quality, tool calls and multiple concurrent agents. DeepSeek's cloud API can support a preliminary experiment, but that experiment is not yet an on-prem deployment
Yes, this connection can be included in a corporate pilot. Yandex AI Studio provides model APIs and states that its models and platform components are hosted in Russia. We select a model and verify compatibility with the agent engine and Guild tools. This is a cloud API in Russia and requires access to Yandex. To meet your overall hosting requirements, we separately agree on Guild services and other integrations
Yes. If your policies permit external connections, Enterprise can support an agreed configuration of Claude Code and Codex with corporate Anthropic and OpenAI APIs. Accounts and payment remain with you. We verify API profiles and their compatibility with Guild during the pilot; this configuration needs access to the selected providers
The agent engine works with files and tools, while the model generates answers and decisions. Connecting Claude Code through a compatible API gateway is one option to investigate. Anthropic does not support routing Claude Code to third-party models, so compatibility and commercial distribution terms require separate review. The pilot determines the final engine configuration
In the proposed architecture, the model runs on a separate inference server, while Guild and the agent run on the user's computer. Model weights do not need to be stored or run on every laptop. Server requirements depend on the model, context size and number of concurrent agents; we establish them through measurements
The isolated pilot includes internal activation and licence checks, hosting for the account portal and relay, and an agreed update delivery method. Agents retain the tools available inside your network. We document all components and network connections before testing isolation
macOS — available; Windows — coming soon; Linux — coming soon; iOS — available; Android — coming soon; Web version — available. We plan the pilot around released apps. The mobile remote needs access to your company network, and the agent computer needs power and connectivity. Phone access and notifications are tested separately: external push services are not part of a fully isolated configuration
Cost and timing depend on the task, model, infrastructure and customisation required. You receive task and isolation test results, load measurements, an agreed installation configuration and terms for the next stage. Server capacity, component licences and support are considered when preparing the proposal
Start with your task
Your next AI team — inside your company
Tell us about your task, infrastructure and isolation requirements. Together we will define the model, pilot scope and success criteria
Your email app will open. You can also write directly to: guild@morozov.digital