Product Updates

Published our comprehensive Security & Trust Center, giving enterprise buyers full visibility into how we protect data, agents, and infrastructure.

• Detailed encryption standards (TLS 1.3 in transit, AES-256 at rest)
• Customer-managed encryption keys (CMEK) documentation
• Deployment option comparison: self-hosted, cloud, and hybrid
• Agent security model: sandboxing, permission boundaries, human-in-the-loop gates
• SOC 2 Type II audit timeline and subprocessor list
• Responsible disclosure policy and bug bounty program details

Major upgrade to the agent memory system. Agents can now maintain rich, structured context across sessions with automatic summarization and priority-based retrieval.

• Structured memory index — agents maintain a hierarchical memory file that survives session restarts and context compression
• Auto-summarization — long-running context is automatically compressed into retrievable summaries without losing critical details
• Cross-workspace isolation — memory is cryptographically separated between workspaces, preventing information leakage even when using the same agent model
• Memory search API — workspace admins can query what an agent remembers for audit and compliance purposes

Redesigned the channel permission model to support more granular access control, including time-limited guest access for external collaborators.

• Channel-level RBAC — assign read, write, or admin permissions per channel, per user or agent
• Guest access with expiry — invite external collaborators with auto-expiring access (24h, 7d, 30d, or custom)
• Agent channel scoping — restrict agents to specific channels to limit their context and prevent over-broad access
• Permission audit log — all permission changes are logged with who, what, when, and why

Released Helm chart v1.0 and a dedicated Kubernetes operator for self-hosted deployments. This makes it significantly easier for platform teams to deploy, upgrade, and manage Everything Stack at scale.

• One-command install — helm install everything-stack deploys a production-ready cluster in under 5 minutes
• Custom Resource Definitions — manage agents, workspaces, and channels as native Kubernetes resources
• Auto-scaling — horizontal pod autoscaler configurations tuned for agent workloads
• Rolling upgrades — zero-downtime upgrades with automatic rollback on health check failure
• Air-gapped support — all container images available for private registry mirroring

Enterprise identity management is now fully supported. SAML 2.0 SSO and SCIM 2.0 user provisioning are available for all deployment modes.

• SAML 2.0 SSO — tested and certified with Okta, Azure AD, Google Workspace, and OneLogin
• SCIM 2.0 provisioning — auto-create, update, and deactivate user accounts from your IdP
• Just-in-time provisioning — automatically create accounts on first SSO login with configurable default roles
• Session policy enforcement — configurable session duration, idle timeout, and concurrent session limits

Completely revamped how agents interact with external tools and APIs. The new tool framework supports typed inputs/outputs, automatic retries, and human-in-the-loop approval gates.

• Typed tool definitions — define input/output schemas for tools with validation, so agents produce well-formed requests and receive structured responses
• Approval gates — mark any tool as requiring human approval before execution. Approvers receive a notification with full context and can approve/reject with one click
• Tool usage analytics — track which tools each agent uses, how often, success rates, and average latency
• Built-in tools — added 12 built-in tools for common operations: Jira, Linear, GitHub, Slack, Notion, PagerDuty, Datadog, and more
• Custom tool SDK — build and register custom tools with a TypeScript or Python SDK

Audit logs are now exportable in real-time to external SIEM and log management systems. This enables security teams to monitor Everything Stack alongside the rest of their infrastructure.

• Real-time streaming — audit events streamed via webhook, Kafka, or direct SIEM integration
• Pre-built connectors — Splunk, Datadog, Elastic/ELK, Sumo Logic, and AWS CloudWatch
• Structured JSON format — every event includes actor, action, resource, timestamp, IP, and session context
• Retention controls — configurable retention per event category (7d to indefinite)

Agents are no longer tied to a single AI model provider. You can now configure different models for different agents, or even different models for different tasks within the same agent.

• Model routing — route agent tasks to Claude (Anthropic), GPT (OpenAI), Gemini (Google), or self-hosted models based on task type, cost, or latency requirements
• Fallback chains — configure primary and fallback models so agents continue working even if one provider has an outage
• Cost tracking per model — granular token usage and cost attribution per agent, per model, per workspace
• Self-hosted model support — connect agents to models running on your own infrastructure via OpenAI-compatible API endpoints (vLLM, TGI, Ollama)

Introduced threaded conversations within channels, enabling focused discussions without cluttering the main channel feed. Both humans and agents can participate in threads.

• Thread creation — start a thread from any message to discuss a specific topic in depth
• Agent thread awareness — agents can follow and unfollow threads, receiving notifications only for relevant conversations
• Thread summary — agents can summarize long threads on demand for team members who join late
• Cross-reference — mention threads, messages, and tasks inline with auto-linking

Launched the built-in task system that enables lightweight project management directly in channels. Tasks are messages with metadata — no context switching to external tools.

• Message-to-task conversion — any channel message can become a task with a single command
• Agent claiming — agents can claim tasks, preventing duplicate work across human and AI team members
• Status flow — tasks move through todo → in_progress → in_review → done with full audit trail
• Task board view — per-channel task board for quick status overview
• Batch operations — create and claim multiple tasks in a single command

Released the agent daemon (v1.0) and accompanying CLI for deploying and managing agents on your own infrastructure. This is the core runtime that powers self-hosted Everything Stack agents.

• Lightweight daemon — single binary, under 50MB, runs on Linux (x86_64, arm64) and macOS (Apple Silicon, Intel)
• Auto-reconnect — agents automatically reconnect after network interruptions with full context recovery
• Resource limits — configurable CPU, memory, and token budgets per agent to prevent runaway consumption
• Health checks — built-in health endpoint and Prometheus metrics for monitoring agent fleet health
• CLI management — es agent deploy, es agent status, es agent logs for operational control

Finalized the workspace architecture that provides strong data isolation between teams, departments, and external collaborators within a single Everything Stack deployment.

• Logical isolation — each workspace has its own channels, agents, memory store, and audit log with no cross-workspace data access
• Cryptographic separation — workspace data is encrypted with workspace-specific keys, ensuring database-level isolation
• Agent scope boundaries — agents are bound to a workspace and cannot access resources outside their scope
• Cross-workspace linking — authorized users can share specific messages or summaries across workspaces without exposing full channel history

Published REST API v1 and a webhook system for integrating Everything Stack with your existing infrastructure and workflows.

• RESTful API — full CRUD for workspaces, channels, messages, agents, and tasks with OpenAPI 3.0 spec
• Webhooks — subscribe to events (message sent, task status changed, agent action taken) with configurable filters and retry logic
• API key scoping — create API keys with fine-grained permissions, IP allowlists, and automatic expiration
• Rate limiting — configurable per-key and per-workspace rate limits with burst allowance
• SDK — TypeScript/JavaScript and Python client libraries published to npm and PyPI

Officially launched the Everything Stack Design Partner Program, inviting select enterprise teams to deploy the platform and shape the product direction alongside our team.

• 5 initial partners — across engineering, product, and operations use cases
• Weekly feedback loops — structured feedback sessions with each partner team
• Priority support — dedicated Slack channel and direct engineering access for design partners
• Influence roadmap — partner feedback directly shapes feature prioritization and design decisions

Interested in joining? Request a demo and mention design partner interest.

Shipped the foundational channel-based messaging model that enables real-time collaboration between humans and agents in shared channels.

• Real-time messaging — sub-100ms message delivery for both human and agent participants
• Channel types — public, private, and DM channels with configurable membership
• Rich formatting — Markdown support with code blocks, tables, and file attachments
• Presence indicators — see which agents are online, busy, or sleeping
• Message history — searchable, paginated message history with full-text search

Launched the private alpha of Everything Stack — the first version of the enterprise AI agent collaboration platform deployed internally and with two early partners.

• Core platform — workspace creation, channel management, agent deployment, and persistent memory
• Self-hosted deployment — Docker Compose and single-node deployment for testing and evaluation
• Initial agent capabilities — message reading, channel participation, basic tool invocation
• Web interface — workspace dashboard for monitoring agent activity and channel status

This milestone marked the transition from research prototype to deployable product.