Talos
Talos is an autonomous AI agent built for production teams running real operational work.
Deploy an always-on agent runtime with deep integrations, shared multi-channel context, configurable approvals, and durable memory. Talos combines active behavior, parallel agent execution, and enterprise controls to keep your business workflows moving 24/7.
Key benefits
24/7 autonomous operations
Talos runs continuously with heartbeat monitoring, proactive checks, and persistent execution.
Multi-channel continuity
Conversations on iMessage, Signal, Slack, and other channels all share the same memory context.
Model-flexible reasoning
Optimized for Claude Opus 4.6 with policy-based routing across multiple model providers.
Hybrid memory retrieval
Full-text plus vector search with adaptive reranking keeps context relevant as workloads evolve.
Deep system integrations
Email, calendar, GitHub, Notion, browser automation, SSH, and filesystem tasks are first-class.
Approval-aware automation
Configure exactly which actions require human approval and which can execute automatically.
Parallel sub-agent execution
Spawn isolated background sub-agents for parallel tasks without blocking primary workflows.
Scoped memory isolation
Separate memory contexts by project or workstream to reduce cross-talk and enforce boundaries.
Audit-ready by default
Typed memory, decision lineage, and replayable history make every action traceable.
How Talos works
Talos runs as continuous operational infrastructure. Each session starts with fresh context, executes against scoped memory and connected systems, and records durable state so work can continue across shifts, channels, and runtimes.
- Briefing-first startup: Talos receives the last 48 hours of relevant context and active thread cards at session start.
- Policy-driven execution: Approval workflows determine what runs autonomously and what pauses for human review.
- Smart model routing: Routine tasks can use lower-cost models while high-stakes reasoning escalates to Claude Opus 4.6.
- Parallel operations: Background sub-agents run isolated tasks in parallel and synchronize through session management and shared memory.
Communication channels
Talos operates where your team already communicates. Channel selection changes the interface, not the memory.
Supported channels
- iMessage
- Signal
- Telegram
- Discord
- Slack
- Google Chat
Shared context across all channels
- Start work in one channel and continue in another with full continuity.
- Thread cards and memory scopes preserve per-workstream context.
- Proactive status updates route to the right people without context loss.
Memory system
XMDB memory is designed for ongoing operational work, not short chat windows. Retrieval quality improves over time, and context is isolated by scope to support serious production deployments.
Hybrid search
Combines full-text retrieval with vector embeddings using ONNX and mxbai-embed-large.
Typed memory entries
Memory is structured as notes, decisions, tasks, events, and insights.
Contrastive reranker
A 6-signal scoring layer adapts over time to improve retrieval precision in production.
Thread cards
Per-workstream state tracking keeps active work organized and resumable.
48-hour briefings
Each session starts with the most relevant context from the prior 48 hours of activity.
Self-improvement flywheel
Automated evals identify retrieval gaps and continuously improve memory ranking quality.
Scope isolation
Independent memory scopes per project or workstream protect boundaries and reduce noise.
Model support
Talos is optimized for Claude Opus 4.6 (Anthropic) as the primary reasoning engine for complex operational decisions.
- Also supports OpenAI (GPT-4o, o1), Google Gemini, MiniMax, and local models via Ollama.
- Bring your own API keys or use XMDB-managed model access.
- Model policy can be scoped per workflow, team, or action type.
Smart routing
Cost-aware by default
Routine tasks can route to lower-cost models like Haiku or Sonnet, while complex reasoning and higher-risk decisions escalate to Opus automatically.
Connectors and integrations
Talos executes real operations by connecting directly to the systems your team already depends on.
Inbox monitoring, triage, and outbound message sending.
Calendar
iCloud and Google Calendar sync for scheduling and follow-up workflows.
GitHub
Issue triage, pull request workflows, and CI-aware actions.
Notion
Read and update pages, databases, and structured blocks.
Browser automation
Headless and interactive browsing, screenshots, and web scraping.
SSH
Secure remote execution on infrastructure and operational hosts.
Filesystem
Read, write, and organize files across local and connected environments.
Scheduled jobs
Cron-based recurring tasks and operational maintenance routines.
Camera and screen capture
Capture visual context from paired devices when workflows require it.
TTS voice output
ElevenLabs output for spoken updates and agent-generated responses.
Web search
Brave API search integration for research and real-time information retrieval.
Social publishing
Bluesky and related social posting workflows with approval controls.
Deployment options
Choose the operating model that matches your environment and compliance requirements.
Cloud (VPS)
Always-on runtime with straightforward scaling for growing agent workloads.
Local (Mac mini/studio)
macOS-native automation, deep Apple ecosystem support, and low operating cost.
Hybrid (recommended)
Mac mini as the primary runtime with VPS expansion for burst compute and redundancy.
For iCloud-backed workflows, Talos deployments should use a dedicated Apple ID for operational isolation and account hygiene.
Autonomous behavior
Talos is designed to operate continuously and proactively, with clear human controls.
Heartbeat system
Periodic checks for inboxes, calendars, support queues, and social channels.
Proactive outreach
Talos notifies humans when action is required instead of waiting for prompts.
Background sub-agents
Isolated sub-sessions run in parallel for independent tasks and longer workflows.
Approval workflows
Fine-grained control over which actions auto-execute and which pause for review.
Continuous runtime
Operates 24/7 as operational infrastructure, not just on-demand chat.
Agent coordination
Talos coordination has moved beyond prototype patterns to managed multi-agent session orchestration.
Multi-agent sessions
Spawn sub-agents for parallel execution while maintaining shared objectives and state.
Session management
List active sessions, pass messages between sessions, and inspect historical activity.
Agent-to-agent messaging
Native coordination through the toku.agency DM system for structured agent communication.
Beta and experimental features
Teams can opt in to early capabilities focused on memory quality and advanced retrieval behavior.
- Self-improving memory retrieval with ML flywheel tuning
- Contrastive learning reranker improvements and adaptive scoring
- Entity-aware search (currently in development)
- Opt-in early-adopter program for experimental capabilities
Production onboarding flow
Talos onboarding is structured around operating requirements, controls, and deployment readiness.
Step 1
Plan the operating scope
Define workstreams, approval thresholds, and memory scope boundaries per business function.
Step 2
Connect channels and systems
Enable communication channels plus operational integrations like email, calendar, GitHub, and Notion.
Step 3
Deploy and harden policies
Choose Cloud, Local, or Hybrid deployment and configure approval gates for sensitive actions.
Step 4
Go live with continuous tuning
Launch 24/7 operations, monitor heartbeat outcomes, and improve retrieval with the eval flywheel.
Use cases
Customer support operations
Monitor support queues, draft responses, escalate blockers, and keep teams informed in real time.
Revenue and account workflows
Track inbound leads, follow up across channels, and keep CRM-related tasks moving continuously.
Engineering execution
Coordinate issues, PRs, CI status, and release checklists with persistent task memory.
IT and security operations
Run scheduled checks, monitor infrastructure signals, and trigger human review when thresholds are crossed.
Executive operations
Handle inbox triage, calendar prep, meeting follow-ups, and decision tracking across teams.
Content and social publishing
Plan, draft, schedule, and publish content with approval workflows for brand control.
Research and intelligence
Continuously monitor topics, gather sources, and maintain structured insight memory over time.
Back-office automation
Coordinate recurring internal workflows for reporting, reminders, and operational hygiene.
Ship while you sleep, with production controls.
Talos gives your business a continuously operating agent runtime with memory, integrations, approvals, and auditability built in.