Track the evolution of the TwelveAgents architecture and public clients.
First Public Release (v1.0)
Live Now
After extensive research, development, and architectural engineering, TwelveAgents is officially live.
We built a robust, proprietary AI orchestration layer and a deterministic state engine capable of executing complex, multi-tool reasoning — and today we're deploying that infrastructure to the public, alongside a full roster of twelve specialized agents and our v1.0 client interface.
Core Infrastructure
- Deployment of our Proprietary Orchestration Layer, enabling seamless multi-tool chaining and autonomous task execution in a single request.
- All Twelve Specialized Agents shipped and live — Nexo, Zeta, Lexa, Doca, Anix, Dato, Sino, Pixo, Vixo, Sona, Kado, and Maku, each with its own persona and default toolset from day one.
- Curated Agent Brains integrated as swappable reasoning engines, selectable per agent or mid-conversation with zero loss of context, files, or state.
- User-Controlled Context Injection enabled, letting you decide how many past messages are carried into context per turn — balancing continuity, relevance, and resource usage.
- Deterministic State Engine activated to lock foundational facts, validate structural logic, and prevent multi-turn AI drift.
- Surgical Editing Pipeline live, enforcing precision diffing and version control on code and documents rather than blind re-generation, with structural edits applied via exact-path patching alongside targeted text replacement for maximum reliability.
- Concurrency-Safe Artifact Writes enforced, so an edit is only ever applied against the version it was actually based on — preventing an outdated edit from silently overwriting newer work.
- Native Media Pipeline live, letting agents generate, reference, and chain images, audio, video, and documents across multiple steps in one request — without you manually downloading and re-uploading intermediate files between tools.
- Secure Search, querying an independently-indexed search engine with zero query tracking, so agents can pull current information without your queries being profiled.
- Isolated Code Execution, running generated code inside hardware-level isolated micro-VMs — a stronger security boundary than standard containers — keeping every execution ephemeral and air-gapped from core infrastructure.
- Secure Per-User Cloud Storage deployed, providing every account with a securely isolated 1 GB storage allowance for uploaded and generated files.
Public Client
- Cross-Platform Client launched, with minimal server-side retention used only to sync active room state across your devices.
- Unified Workspace Interface launched, providing a persistent, structurally isolated environment to deploy all twelve agents.
- All Twelve Agents Accessible from any chat, each shipping with its own default persona and toolkit from a shared library of 36+ tools spanning research, writing, code, data, and media.
- Extensible & Reducible Toolsets, letting you add tools an agent doesn't have by default, remove ones you don't need, or disable all tools entirely — reshaping any agent's capabilities per chat, entirely through natural language.
- Natural-Language Media Control, generating, editing, and mixing images, audio, video, and documents through plain conversational prompts, with the underlying orchestration handled entirely server-side.
- In-App Media Playback, with generated images mainly parsed and displayed directly in-app, and most generated audio playable in-app as well; other media — including generated video, PDF, Word, and similar files — typically opens through your device's own apps, with a direct download link always provided.
- Code Authoring Environment deployed, featuring dynamic syntax highlighting, line numbering, and LaTeX mathematical typesetting.
- Tool Chain Flow Visibility, showing the sequence of tools called within a workflow along with each step's input and output, so you can see exactly how a multi-step result was assembled.
- Execution Governance enforced behind the scenes; if a workflow hits an operational limit mid-task, the agent tells you exactly what limit was reached and suggests how to continue.
- Credit Tracker built in, giving a real-time view of remaining balance and per-request cost before you send.
- Prompt Protection available as an optional, on-device layer (off by default) that runs entirely locally and persists your chosen setting across sessions, scanning and redacting sensitive data before it's ever sent.