Your data. Your models. Your agents.

The agent platform your team builds and runs together.

Ground every agent in your company's real data. Build them, share them, orchestrate them — on your cloud, with your model, under your permissions.

Bring your own model Bring your own cloud Run fully local MCP-native EU / US data residency

Your data, your infra. BYOC or run local — your call.

Contextful is local-first by default. Sensitive orgs and regulated teams can go further: bring your own cloud account, or run the entire stack on a laptop or internal cluster — fully air-gapped.

Default

Managed, local-first

Desktop app holds an encrypted Parquet replica and a local Loro CRDT. Our cloud coordinates sync — it is never the source of truth. If we disappear, your data is still on your machines in open formats.

  • Zero-ops onboarding
  • Encrypted sync to your devices
  • Works offline, syncs when online
Self-hosted

Run Fully Local

For air-gapped, regulated, or offline-first teams. Hakiri runs on a laptop or internal cluster, Weaver syncs via self-hosted Durable Objects or direct peer-to-peer. No cloud dependency at all.

  • True air-gapped deployment
  • Local LLM inference (Ollama, llama.cpp, vLLM)
  • Open formats — Parquet, JSON, Markdown

A platform where agents are first-class citizens

Agents your team owns

Build, share, fork, and version agents like code. Every agent is a first-class object on the platform — versioned, permissioned, observable, and yours to run on any model.

Multiplayer by default

Humans and agents share the same workspace, the same context, the same audit log. Hand work between teammates and agents in one thread — with CRDT-backed sync that works offline.

Orchestrate, don't babysit

Compose agents into pipelines, route work between them, and let them call each other through MCP. Capability tokens scope every step — agents see only what their caller is allowed to see.

Open formats, your storage

Parquet on disk, Loro CRDT for sync, SQLite for catalog. Standard tools read your data — DuckDB, Pandas, Polars. No proprietary lock-in.

Typed connectors, one interface

WASM-sandboxed connectors for QuickBooks, Google Workspace, Slack, and 50+ tools. One context API across every source — built for agents, not chatbots.

Run on any model

BYOK for OpenAI, Anthropic, Bedrock. Self-host with Ollama, vLLM, or llama.cpp. Switch models per agent — without rewriting prompts, connectors, or permissions.

What teams build on Contextful

Open-source recipes — agent configurations, prompts, and connector wiring — that you fork, run on your own infra, and own end-to-end.

Month-End Close

An agent that tracks open items, flags discrepancies, and drafts reconciliation notes across your accounting stack.

Audit Prep

An agent that gathers evidence, maps controls to documents, and flags gaps before your auditor finds them.

Expense Policy Review

An agent that reads expenses against your policy doc, flags violations, and drafts approval notes.

Regulatory Change Tracking

An agent that monitors regulatory feeds, maps changes to your obligations, and alerts the right owners.

Vendor Risk Review

An agent that pulls vendor security questionnaires, compares against your requirements, and surfaces gaps.

Contract Renewal

An agent that monitors upcoming renewals, surfaces relevant context, and drafts review summaries.

From connectors to a team of agents in five minutes

1

Connect your sources

OAuth into QuickBooks, Google Workspace, Slack, and 50+ tools. Typed WASM connectors land your data as encrypted Parquet on disk.

2

Build & share agents

Fork a recipe or write your own. Version it, scope it with capability tokens, and share it with the team. Any MCP client — Claude Code, Cursor, your own — connects through the same interface.

3

Collaborate & orchestrate

Run agents alongside teammates in shared workspaces. Compose them into pipelines, hand off work, and watch every action land in the audit log with full provenance.

How is Contextful different from Glean, Dust, or Notion AI?

Most agent products are search boxes or chat surfaces over your SaaS data. Contextful is the platform layer underneath — the connectors your agents can write, the governance your auditors can verify, the data you can export today.

Dimension Glean Dust Notion AI Contextful
Core metaphor Enterprise search + assistant Chat-first AI assistants AI inside Notion docs Platform for agents your team builds
Where data lives Glean cloud (SaaS) Dust cloud (SaaS) Notion cloud (SaaS) Local-first; BYOC or self-host; Parquet on disk
Custom connectors Vendor-built; SDK on roadmap Vendor-built catalog Limited to Notion + a few apps Agent-authored, typed contracts, hot-swappable
When upstream APIs change File a ticket, wait File a ticket, wait N/A Self-healing — agents propose patches the same day
Permission model Inherits source ACLs in Glean's index Per-agent scopes in Dust's cloud Notion page permissions Capability tokens scoped to org chart, per call
Bring your own model Glean-managed Multi-provider, hosted Notion-managed BYOK — OpenAI, Anthropic, Bedrock, Ollama, vLLM
Data export Limited; via support Limited; via support Notion export One command, today — Parquet + SQLite, open schema
Buyer fit Enterprise ($30+/seat, 6-figure deals) Mid-market & enterprise Anyone on Notion Regulated SMBs & teams that need their own perimeter

Comparison reflects publicly documented capabilities as of 2026-06. Vendors evolve — your mileage will too.

Pricing for the platform — pick how you deploy

Starter

$25/seat/month

  • 5 connectors
  • Local-first sync
  • 100k synced records
  • MCP endpoint included

Enterprise

Custom 

Run the platform inside your perimeter. We help your team stand it up — connectors, IAM wiring, compliance posture — then hand back the keys.

  • Everything in Team
  • BYOC + on-prem / air-gapped deployment
  • Custom connectors & SSO
  • SOC 2 report, DPA, deployment support
Talk to us