1 Definitions

For purposes of this Services Description, the following definitions apply:

• "Customer" — The individual or organization that has entered into an agreement with DeepCerebra for access to the Services.
• "User" — Any individual who accesses or uses the Services under a Customer's account.
• "Agent" — The DeepCerebra Coder AI system that plans, edits code across multiple files, runs commands, and calls tools on the User's behalf.
• "Your Content" — Source code, files, prompts, and other materials submitted to, stored in, or generated with the Services.
• "Generated Output" — Code, edits, documents, and other content produced by the Agent.
• "ZDR LLM API" — DeepCerebra's Zero Data Retention gateway for routing requests to large language model providers.
• "BYOK" — "Bring Your Own Key," the option for a User to supply their own model provider credentials.

2 DeepCerebra Coder Services

DeepCerebra Coder is an agentic, multi-file software-development assistant available as a web application, desktop applications (Windows, Linux, macOS), a command-line interface, and an API. It helps Users plan, write, refactor, debug, and understand code across an entire project.

2.1 Surfaces

Web Application: A browser-based workspace with an explorer, editor, integrated terminal, diff/apply review, and chat. Project files are held in confidential in-browser storage.

Desktop Applications: Native installers for Windows, Linux, and macOS, with access to the local file system and the ability to run local model runtimes.

CLI & API: Paid Programmatic access for scripting, CI/CD, and integrations, authenticated with a personal access token.

2.2 Core Capabilities

3 Advanced Features

3.1 Tiered, Adaptive Model Routing

When the model selector is set to automatic, DeepCerebra routes simple tasks to a fast, economical model and complex or planning-heavy tasks to a stronger model, balancing cost and quality. Users may also select a specific model.

3.2 Model Context Protocol (MCP) Servers

Users can connect MCP servers to extend the Agent with external tools and data sources, configured through an mcp.json settings file.

3.3 Agent Hooks, Steering & Skills

Users can define agent hooks (file, contextual, task, or manual triggers), steering documents that describe product, structure, and technology conventions, and reusable skills that the Agent applies automatically.

3.4 Git-Backed Explorer & Collaboration

Users can load a project directly from a Git repository (GitHub or GitLab), with checkout, update, commit, and merge workflows, enabling collaborative team development through DeepCerebra.

3.5 Local GPU and Local Models

Users can run local models via Ollama or LM Studio on a local or networked GPU, in both the desktop and web applications, for private, offline, or cost-controlled inference.

3.6 Folder-Scoped Chat Sessions

Chat history is scoped to the open project folder; changing folders starts a new session and loads that folder's prior sessions so Users can resume where they left off.

4 Plans and Feature Access

DeepCerebra Coder is offered through subscription plans that include a monthly allowance of AI usage credits (denominated in U.S. dollars), with the ability to purchase additional usage credits. Users may also reduce or avoid metered usage by bringing their own model keys (BYOK) or running local models.

Capability	Basic	Pro	Ultimate
Monthly price	$20	$75	$200
Included monthly AI usage	$10	$30	$150
Agentic chat & multi-file edits	✅	✅	✅
Integrated terminal & diffs	✅	✅	✅
BYOK & local models	✅	✅	✅
MCP servers, hooks, steering & skills	✅	✅	✅
Git-backed collaboration	✅	✅	✅
CLI & API access	✅	✅	✅
Priority support	❌	✅	✅

Specific included allowances, rates, and plan features are shown in your account dashboard and may be updated from time to time as described in the Terms of Service.

5 Usage Metering and Fair Use

AI usage is metered based on the tokens consumed by the underlying models and the published rates. Cached inputs may be billed at a reduced rate. DeepCerebra applies per-turn budget safeguards and cost controls to prevent runaway usage. Customers are responsible for usage incurred under their account, including usage by the Agent when executing authorized workflows.

6 Compliant Use

Customers are solely responsible for their and their Users' access to and use of the Services, and shall ensure that such use complies with all applicable laws, including those relating to intellectual property, privacy, and export control. Customers are responsible for reviewing and testing Generated Output before relying on or deploying it, and for the actions they authorize the Agent to take in their environments.

7 Technical Requirements

The web application requires a modern browser (Chrome, Edge, Firefox, or Safari) with current versions recommended. Desktop applications are provided for Windows, Linux, and macOS. Using local models requires a compatible local runtime (Ollama or LM Studio) and sufficient local or networked GPU resources. A stable internet connection is required for cloud model routing, account, and billing features.