Compliance Framework

Context Pilot does not act as a data controller or data processor as defined under GDPR. No personal data is collected, stored, or processed on Context Pilot infrastructure (which does not exist). The operator is the sole data controller for any data processed locally on their workstation.

• Data minimization: Satisfied by architecture. No data is collected beyond what the operator explicitly provides to their configured LLM provider.
• Purpose limitation: All data processing serves the operator's immediate coding assistance needs. No secondary use occurs.
• Storage limitation: All data resides on the operator's filesystem under their full control. No external retention.
• Data subject rights: Exercisable through direct filesystem access. See the Data Privacy page for details.
• Third-party transfers: LLM API calls may constitute a data transfer to the provider's jurisdiction. This is governed by the operator's agreement with their chosen provider, not by Context Pilot.

SOC 2 is designed for service organizations that process, store, or transmit customer data through cloud infrastructure. Context Pilot does not operate as a service organization: there is no cloud service, no customer data processing, and no multi-tenant infrastructure. A SOC 2 engagement would be structurally inapplicable.

However, the trust service criteria addressed by SOC 2 are substantially satisfied through Context Pilot's architectural controls:

• Security: Local-first architecture, no network attack surface, path traversal protections, SSE ticket authentication
• Availability: No centralized service to suffer outages; each deployment is independent
• Processing Integrity: SHA-256 hash chain for critical files, 1001 enforced lint rules, automated CI
• Confidentiality: No data collection infrastructure; credentials isolated per-provider
• Privacy: No telemetry, no analytics, no user tracking; all data local-only

Daharness has not pursued ISO 27001 certification as it is designed for organizations operating information security management systems for cloud services. The project implements controls aligned with ISO 27001 Annex A categories where applicable to the open-source engine it runs:

• A.8 Asset Management: All code assets tracked in Git with signed commits and CI enforcement
• A.9 Access Control: Filesystem-scoped access, credential isolation, SSE ticket system
• A.12 Operations Security: Automated build pipelines, dependency scanning (Dependabot), CodeQL analysis
• A.14 System Development: 1001 lint rules, structural enforcement (500-line/8-entry limits), hash chain integrity
• A.16 Incident Management: Responsible disclosure program via GitHub Security Advisories

Daharness is a general-purpose coding assistant. It does not process Protected Health Information (PHI) by design. If an operator uses Daharness in a healthcare context, the operator is responsible for ensuring that no PHI is transmitted to LLM providers without appropriate Business Associate Agreements (BAAs) in place with those providers.

Context Pilot does not sell, share, or collect personal information as defined under CCPA. The software does not have the technical capability to collect consumer data. No opt-out mechanism is required because no data collection occurs.

The AGPL-3.0 grants the right to use, study, modify and redistribute the software, including commercially. Those rights carry obligations:

• Distributing the software, modified or not, requires making the corresponding source available to the recipient under the same licence
• Derivative and combined works fall under the same terms. The licence is strong copyleft, not permissive, and does not allow redistribution under proprietary terms
• Section 13 extends that obligation to network use: where users interact with a modified version remotely over a network, they must be offered the corresponding source of that version
• Copyright and licence notices, and a record of modifications, must be preserved

The licence includes a standard disclaimer of warranty and limitation of liability. The authoritative text is the LICENSE file in the engine repository.

Not addressed here: how section 13 applies to Daharness on our servers, which is a hosted service built on this engine. This document does not answer that question and makes no undertaking about it.

All direct dependencies use permissive open-source licenses (MIT, Apache 2.0, BSD). No copyleft (GPL, AGPL) or proprietary dependencies exist in the dependency tree. The project's Cargo.lock and pnpm-lock.yaml files provide reproducible dependency resolution for independent audit.

Under the EU AI Act risk classification framework, Daharness is categorized as a low-risk or minimal-risk AI system. It does not perform biometric identification, critical infrastructure management, education scoring, employment filtering, law enforcement, or any other high-risk use case enumerated in Annex III. Daharness is a general-purpose coding assistant that operates under direct human supervision at all times.

• Human oversight: All AI outputs require explicit human review. The tool does not make autonomous decisions affecting natural persons.
• Transparency: The system is clearly identified as an AI assistant. Users interact with it knowingly and intentionally.
• No prohibited practices: No subliminal manipulation, social scoring, or real-time biometric identification capabilities exist.
• General-purpose AI model: Daharness does not train or fine-tune AI models. It uses third-party provider APIs for inference only. GPAI obligations apply to the model providers (Anthropic, OpenAI, etc.), not to Daharness.

The design of Daharness aligns with the NIST AI RMF's core functions:

• Govern: Open-source development with public code review, documented architecture, and transparent decision-making.
• Map: Clear system boundaries — Daharness is a developer tool, not an autonomous decision system. Usage context is well-defined.
• Measure: All AI interactions are logged locally (conversation history) for operator review. Cost tracking provides usage visibility.
• Manage: Guard rails (cost limits, message limits, duration limits) prevent runaway AI operations. Human approval is required for destructive actions.

Daharness implements multiple safeguards for responsible AI use:

• Cost controls: Per-agent budget limits prevent uncontrolled API spending. Circuit breakers halt operations when budgets are exceeded.
• Audit trail: Complete conversation history, tool call logs, and decision traces are preserved locally for operator review.
• Model transparency: The operator selects and configures the LLM model. Model identity is always visible. No hidden model switching occurs.
• Output attribution: AI-generated content is clearly distinguished from user input in the conversation interface.
• Operator control: The operator can interrupt, redirect, or terminate AI operations at any time through guard rails, keyboard shortcuts, or process termination.

The Context Pilot engine uses only standard, commercially available cryptographic implementations (TLS via OpenSSL/rustls, SHA-256 via the sha2 crate). No custom or novel cryptographic algorithms are implemented. The engine falls under EAR Exception TSU (Technology and Software Unrestricted) for publicly available encryption source code. As open-source software published on GitHub, it qualifies for the publicly available exemption under both US EAR and EU Dual-Use Regulation.

The Context Pilot engine is distributed as open-source software under the GNU Affero General Public License v3.0. Distribution is subject to GitHub's terms of service and applicable US export control laws. The software itself does not interact with financial systems, payment processors, or sanctioned entities. Operators are responsible for ensuring their use of third-party API providers complies with applicable sanctions in their jurisdiction.

The web frontend uses semantic HTML, ARIA attributes, and keyboard-navigable controls. The interface is built with React and shadcn/ui components that include accessibility foundations. Current accessibility features include:

• Semantic HTML structure with proper heading hierarchy
• Keyboard navigation for all interactive elements
• Sufficient color contrast ratios for text and interactive elements
• Screen reader compatibility for core workflows
• No motion-sensitive content or auto-playing media

Formal WCAG 2.1 AA compliance audit has not yet been conducted. Accessibility improvements are accepted as contributions and prioritized alongside feature development.

The terminal user interface (TUI) renders through standard terminal emulators and is compatible with terminal-based screen readers. All interactive elements are keyboard-driven by design (the TUI has no mouse interaction as its primary mode). Color is used for enhancement only — all information is also conveyed through text labels and structural formatting.