Headmaster is GCAP's persistent AI workforce layer for organizations that need memory, tools, workflows, approvals, channels, automations, and model routing in one governed product.
The homepage stays broad. This page holds the deeper product details, including integrations configuration and the model stack.
The dashboard gives teams a shared view of what Headmaster is doing, what needs attention, and which workflows are active across the workspace.
Chat is the front door. Headmaster can load context, use approved tools, delegate to agents, and prepare the next artifact for review.
Teams can package recurring work into skills and workflows so Headmaster does not start from scratch every time.
Guided runs make execution visible: inputs, steps, tools, draft output, and approval checkpoints are structured instead of hidden.
Headmaster can retain approved preferences, formats, decisions, and working context so teams do not have to restate the same instructions.
Documents and knowledge sources give Headmaster the source material it needs for reports, drafts, summaries, and decisions.
Specialist profiles help split work across research, writing, analysis, operations, and technical execution without losing central control.
Recurring reports, summaries, reminders, and operational checks can run on schedule with review gates where the organization needs them.
Approvals keep people in control of external communications, sensitive edits, final documents, and important operational actions.
Execution history gives teams a practical record of workflow status, agent involvement, durations, and outcomes.
The deeper integrations layer belongs on the product page: channels, tools, connectors, MCP servers, webhooks, API keys, and provider configuration.
Headmaster is model-agnostic. It can use cloud models, coding models, local models, enterprise endpoints, custom endpoints, and TayX.
Telegram, Discord, Slack, WhatsApp, Signal, iMessage (BlueBubbles), WeChat, WeCom, QQBot, Matrix, Email, SMS, DingTalk, Feishu, Mattermost, Home Assistant, Webhooks. One agent. Every surface.
Grouped as: Messaging, Enterprise, Dev & Infra
Anthropic Claude, OpenAI GPT-5.5, Google Gemini, AWS Bedrock, NVIDIA NIM, xAI Grok, Kimi K2.6, Xiaomi MiMo, Ollama (local), OpenRouter, Vercel AI Gateway, Arcee AI, and more.
Live model switching mid-session. Fast Mode for priority processing.
Isolated subagents with their own conversations, terminals, and Python RPC scripts. Orchestrator role can spawn its own workers.
Configurable spawn depth. Concurrent siblings share filesystem state without conflicts.
Local, Docker, SSH, Singularity, Modal. Container hardening and namespace isolation.
Real isolation, not a shell wrapper.
Register slash commands, dispatch tools, veto tool execution pre-call, rewrite tool results, transform terminal output, add custom dashboard tabs, ship custom image gen backends.
Full lifecycle hooks.
Inject a note the running agent sees after its next tool call, without interrupting the turn or breaking prompt cache.
Course-correct in flight.
Priority processing for OpenAI and Anthropic models. Toggle /fast to route through priority queues.
Significantly lower latency.
Full backup and restore of configuration, sessions, skills, and memory. Migrate between machines or snapshot before major changes.
Start with one workflow, connect the right tools, and let your AI workforce learn from real operations.