Skip to main content
Platform Overview

The platform behind the coworker.

Context, action, governance, and memory — one runtime that turns 18 data sources and 100+ tools into a teammate you can delegate to, audit, and trust.

18 Connectors·100+ Agent Tools·4 Trigger Types·4 Surfaces·T0–T5 Risk Tiers

Architecture

From your sources to every surface.

Data flows in permission-aware, work flows out approval-gated.

18 Sources
Confluence · Slack · GitHub · S3 · SQL
ZenSearch Core
Context · Agents · Governance · Memory
4 Surfaces
Slack · Teams · Web · Extension

Pillar 01 — Context

It knows before it acts.

Every answer and every action is grounded in retrieval: hybrid vector search with reranking, verified citations, and confidence scoring — over exactly the documents each person is allowed to see.

  • Dense + sparse hybrid vector search
  • Cross-encoder reranking for top-K precision
  • Verified citations with page/slide/line references
  • Answer confidence scoring (High / Medium / Low)
  • Query expansion and shape-aware retrieval
  • Permission-aware retrieval — people only see what they're allowed to

Pillar 02 — Action

Plans the job, works the tools, finishes.

A graph-based agent runtime executes every run the same way, whether it started from a chat message or a 2 AM webhook.

Init → Provision → CostCheck → Plan → LLM ⇄ Tools → Synthesize → Complete

100+ built-in tools

Search, SQL databases, Google Workspace, Microsoft 365, Zendesk, ServiceNow, HubSpot, custom webhooks, and MCP servers.

Multi-agent delegation

Agents hand subtasks to specialist agents on the same team — depth-limited, budget-clamped, with costs rolled up to the parent run.

Three-layer memory

Persistent facts and preferences, observational compression on long conversations, and procedures learned from successful runs.

Canvas artifacts

Versioned documents, code, and data the agent drafts and revises — with diff view and user editing.

Automations

Scheduled and triggered agent runs with delivery to webhook, Slack, or email — and auto-resume when a run pauses on a budget.

Budgets & recovery

Pre-flight cost gates, per-tool timeouts, mid-run cost ceilings, transient-failure retries, and pause-and-resume checkpoints.

Pillar 03 — Triggers

Give it work the way you'd give a person work.

Mention it, email it, schedule it, or let your systems call it — every trigger lands in the same governed runtime.

Chat

Ask in Slack, Teams, or the web — switch to research mode for deep multi-step work.

Email

Give an automation its own address. Forward it work; replying to its emails resumes the run.

Schedule

Cron-scheduled runs with history, verification contracts, and bounded auto-resume.

Events

Webhooks from Jira, Zendesk, GitHub, Confluence, Salesforce, and HubSpot — attributed to the triggering user.

Meetings

Zoom, Google Meet, and Teams events kick off runs — transcripts optionally ingested.

Pillar 04 — Governance

Autonomy with a paper trail.

Every tool carries a risk tier; every consequential action waits for the right human; every run leaves a trace you can audit.

  • Risk tiers T0–T5 with per-team policy overrides
  • Approvals decided from Slack, email reply, or the web — routed through your org chart and on-call schedules
  • Per-run and per-team-day cost ceilings with live cost metering
  • Identity federation (Slack, Teams, Google, OIDC, SAML)
  • Memory scope, retention, and visibility policies
  • Complete audit logging with full reasoning traces

Pillar 05 — Memory

Better in month two than on day one.

Three memory layers compound: what it's told, what it observes, and what it learns by doing.

01

Facts & preferences

Durable team knowledge the agent saves and recalls mid-run — rate-limited and deduplicated.

02

Observational compression

Long conversations distill into compact observations — 80%+ token compression on 50-message sessions, so context stays cheap.

03

Procedures

Successful runs become named, reusable playbooks. Admins curate the library: publish, version, and scope by role.

Operations

Control Tower.

One dashboard for everything your coworker does: connector health, model spend, agent performance, schedules, and failures.

Connector health
Model cost tracking
Agent performance
Automation schedules
Failed run triage
Sync monitoring

Deployment

Your infrastructure, your models.

Every model call routes through one gateway — swap providers without touching a single agent, or run fully local in air-gapped environments.

OpenAIAnthropicCohereGroqOpenRouterAzure AI FoundryAmazon BedrockOllamaLM Studio

Get Started

Put it
to work.

Talk to our team about a deployment shaped to your data, identity stack, and compliance program.