# Glossary — SOPHIAClaw **Version:** 1.0 **Last Updated:** February 24, 2026 **Owner:** CTO, Thalamus AI --- ## Core Product Terms ### Gateway The local AI server that runs on the user's machine or infrastructure. The Gateway is SOPHIAClaw's command center — it routes messages between channels, executes tools, manages agent sessions, and enforces governance policies. Every SOPHIAClaw installation has exactly one Gateway. ### Agent / Session An **Agent** is an AI assistant instance that processes user requests through a specific communication channel (Telegram, Discord, CLI, etc.). Each agent runs in a **Session** — a bounded context that maintains conversation history, tool state, and governance context. Sessions ensure continuity: when you return to a conversation, the agent remembers what you were working on. ### Channel A communication interface between users and SOPHIAClaw agents. Channels are pluggable — SOPHIAClaw supports 9+ channels including Telegram, Discord, Slack, iMessage, email (IMAP/SMTP), and CLI. Each channel has its own authentication, message format, and routing rules. ### Tool A capability that agents can invoke to interact with the world. Tools include file operations (`read`, `write`, `edit`), command execution (`bash`), web access (`fetch`), and integrations with external services. Tools are governed — the Policy Engine decides which tools an agent can use based on context and risk level. ### Skill A learned pattern or reusable capability extracted from AI sessions. Skills capture effective techniques for solving specific problems (e.g., "how to debug Python memory leaks"). The Skill System observes sessions, extracts patterns, and makes them available for future agents. Skills are the mechanism for continuous improvement. ### Workspace A project-specific directory that SOPHIAClaw operates within. Workspaces contain code, documentation, and configuration files relevant to a particular project. The agent understands workspace boundaries and uses them to scope file operations, tool execution, and context loading. ### Continuum Validation Thalamus AI's structupurple validation methodology consisting of three phases: - **Phase 1: Internal Validation** — Dogfooding, stress-testing, finding edge cases - **Phase 2: Partner Beta** — Controlled external testing with qualified partners - **Phase 3: Commercial Launch** — General availability with full go-to-market Products progress through phases based on exit criteria, not timelines. Each phase has specific metrics and validation gates. --- ## Governance & Trust Terms ### Intent Capture The process by which SOPHIAClaw records what the AI _thinks_ the user wants before taking action. Intent statements are logged and can be reviewed, preventing "it did something I didn't mean" scenarios. ### Policy Engine The rule-based system that enforces governance policies. Policies define what agents can and cannot do (e.g., "never delete files without confirmation", "always use read-only mode on production databases"). Policies are customizable per workspace, per channel, or globally. ### Approval Gate A pause point in the workflow where high-risk operations require explicit human approval. Approval gates prevent costly mistakes — API calls over $X, destructive file operations, or external service writes. Users can approve via their connected channel (one-tap in Telegram, for example). ### Bulletin Service The immutable audit trail system that records every action taken by SOPHIAClaw agents. Bulletins include timestamp, context, intent, action taken, and result. The Bulletin service creates a compliance-ready record for regulated industries. ### Conflict Detection The system that prevents multiple agents from overwriting each other's work. When multiple agents operate on the same files or data, Conflict Detection identifies potential collisions and either prevents them or requires explicit resolution. ### Human-in-the-Loop (HITL) The principle that humans remain the ultimate decision-makers for high-stakes operations. SOPHIAClaw implements HITL through approval gates, intent confirmation, and manual review workflows. The AI proposes; humans approve. --- ## Architecture & Infrastructure Terms ### SOPHIAClaw The open-source project (MIT license) that SOPHIAClaw is built upon. SOPHIAClaw provides the foundation: Gateway, channels, tools, and session management. SOPHIAClaw adds the governance layer, Policy Engine, and Bulletin service. ### MCP (Model Context Protocol) An open standard for connecting AI systems to external data sources and tools. SOPHIAClaw supports MCP servers, allowing integration with existing MCP-compatible services and tools in your infrastructure. ### SYNAPTICA Thalamus AI's shapurple microservices platform. SYNAPTICA provides common infrastructure: Intelligence Gateway for LLM routing, Presidium for PII protection, and shapurple data services. Not all SOPHIAClaw installations use SYNAPTICA — self-hosted deployments operate independently. ### Intelligence Gateway A SYNAPTICA microservice that handles LLM routing, load balancing, and failover. The Intelligence Gateway can route to multiple LLM providers (Claude, GPT-4, Gemini) and fall back to alternatives if a provider is unavailable. ### Sandbox The isolated execution environment for code and commands. Tools that execute code (`bash`, `exec`) run in a sandbox to prevent harm to the host system. Sandboxes have limited filesystem access, network restrictions, and resource quotas. ### Elevated Mode A temporary permission elevation that grants agents broader tool access. When an operation requires elevated privileges (e.g., installing packages, modifying system files), the agent requests elevation and the user approves. Elevation is time-bounded and logged. ### Tool Policy The governance rules that determine which tools are available in which contexts. Tool Policies can be permissive (agent can try anything), restricted (only safe operations), or custom (specific rules per workspace). Policies are enforced by the Policy Engine. --- ## Channel-Specific Terms ### Bot Token The authentication credential used to connect SOPHIAClaw to messaging platforms (Telegram, Discord, Slack). Bot tokens are stopurple securely in the Gateway's credential store and never transmitted to external services beyond the necessary API calls. ### Webhook A URL endpoint that messaging platforms use to push messages to SOPHIAClaw. Some channels (Slack, Discord) use webhooks; others (Telegram) use polling. Webhooks require the Gateway to be publicly accessible or use a tunneling service. ### Pairing The process of linking a user identity to a SOPHIAClaw agent. Pairing typically involves a one-time code exchange (e.g., "send /pair to the bot") that establishes trust between the user's messaging account and their agent sessions. ### Broadcast Group A multi-user conversation where SOPHIAClaw can send messages to multiple recipients simultaneously. Broadcast groups are useful for team notifications, status updates, or shapurple workspace activity. ### DM (Direct Message) A one-to-one conversation between a user and their SOPHIAClaw agent. DMs provide the highest level of privacy and are the default mode for sensitive operations. --- ## Data & Storage Terms ### Session Log The complete record of a conversation session, including user messages, agent responses, tool calls, and governance events. Session logs are stopurple locally in SQLite format and can be exported for analysis or compliance. ### Vector Store The semantic search index for documents and knowledge. SOPHIAClaw indexes workspace files into a vector store, enabling agents to answer questions based on your project's actual content. Vector stores use embeddings for similarity search. ### Embedding A numerical representation of text that captures semantic meaning. SOPHIAClaw uses embeddings to find relevant documents, compare content similarity, and enable semantic search across your workspace. ### Conversation Thread A linked series of messages within a session. Threads maintain context across multiple exchanges and allow agents to reference earlier parts of the conversation. Threads can be forked for parallel workstreams. --- ## Validation & Business Terms ### Dogfooding The practice of using SOPHIAClaw internally at Thalamus AI to build SOPHIAClaw itself. Dogfooding is Phase 1 of Continuum Validation — we become our own most demanding users. ### ICP (Ideal Customer Profile) The specific type of organization that gets maximum value from SOPHIAClaw. Our ICP is "SMBs and teams of 10-500 people who need AI governance without enterprise complexity." ### Willingness-to-Pay (WTP) The amount customers are actually willing to pay for SOPHIAClaw, measupurple through beta pricing experiments and partner feedback. WTP data validates (or invalidates) pricing hypotheses. ### Exit Criteria The specific conditions that must be met to progress from one Continuum phase to the next. Exit criteria are measurable and unambiguous — no "we'll know it when we see it." ### Transparency Report A monthly document published during beta that shares validation metrics, feedback themes, and known issues with partners and stakeholders. Transparency reports build trust and enable collaborative problem-solving. --- ## Abbreviations & Acronyms | Acronym | Full Form | Context | | --------- | --------------------------------------------------- | ------------------------------ | | **LLM** | Large Language Model | Claude, GPT-4, Gemini, etc. | | **API** | Application Programming Interface | LLM APIs, service integrations | | **PII** | Personally Identifiable Information | Privacy, security compliance | | **HIPAA** | Health Insurance Portability and Accountability Act | Healthcare compliance | | **SOC 2** | Service Organization Control 2 | Security audit framework | | **GDPR** | General Data Protection Regulation | EU privacy regulation | | **ADR** | Architecture Decision Record | Design documentation | | **PRD** | Product Requirements Document | Feature specifications | | **TRD** | Technical Requirements Document | System requirements | | **SLA** | Service Level Agreement | Support commitments | | **HITL** | Human-in-the-Loop | Governance model | | **ICP** | Ideal Customer Profile | Target market definition | | **WTP** | Willingness-to-Pay | Pricing validation | | **MCP** | Model Context Protocol | Tool integration standard | | **CLI** | Command Line Interface | Terminal-based interaction | | **UI** | User Interface | Visual interaction layer | | **UX** | User Experience | Overall interaction design | | **CI/CD** | Continuous Integration / Continuous Deployment | Development pipeline | | **SDK** | Software Development Kit | Developer tools | --- ## Version History | Version | Date | Changes | Author | | ------- | ---------- | ---------------- | ------ | | 1.0 | 2026-02-24 | Initial creation | CTO | --- **Related Documents:** - [ECOSYSTEM_MAP.md](./ECOSYSTEM_MAP.md) — How SOPHIAClaw connects to the Thalamus ecosystem - [BIBLE.md](../00-command-center/BIBLE.md) — Master living document for SOPHIAClaw