# Cork vs Dust: company agents, or coding agents on one repo

<!--
title: "Cork vs Dust: company agents, or coding agents on one repo · Cork"
description: "Cork vs Dust: Dust is a platform for building agents on a company's knowledge and tools, for every team. Cork coordinates the coding agents a team already runs (Claude Code, Codex, Cursor) so they don't overwrite each other on one codebase. Where each wins, and how they work together."
canonical: "https://corkcli.com/blog/cork-vs-dust"
last_updated: "2026-11-16"
published: "2026-11-16"
author: "Kevin Vera"
-->

Cork vs Dust: Dust is a platform for building agents on a company's knowledge and tools, for every team. Cork coordinates the coding agents a team already runs (Claude Code, Codex, Cursor) so they don't overwrite each other on one codebase. Where each wins, and how they work together.

By Kevin Vera · November 16, 2026 · [Comparisons](https://corkcli.com/blog/c/comparisons) · 6 min read

How is Cork different from Dust? They solve different problems. Dust (dust.tt) is a platform where a company builds and runs AI agents on its own knowledge and tools: agents that answer from Notion, Slack and Drive, file tickets, draft replies, run on schedules, and work with people in shared spaces. It is for every team, from sales to support to engineering. Cork works on one problem, for engineering teams: several coding agents (Claude Code, Codex, Cursor) changing the same codebase at once. Cork briefs each one on the code it is about to touch, tells it when another agent changes something it depends on, holds a write against stale code, and turns `git push` into one rebase-test-push step.

Dust can start coding agents. Its Cursor Cloud Agents integration launches them on your GitHub repositories and follows their pull requests. Nothing in Dust coordinates several agents editing one repository. We searched its full documentation for merge conflicts, rebases, pushes, worktrees and renames and found nothing about any of them. Facts about Dust below are from dust.tt and docs.dust.tt, read on 15 November 2026.

## Cork vs Dust at a glance

| Topic | Cork | Dust (as of November 2026) |
|---|---|---|
| What it is | [Coordination layer](https://corkcli.com/blog/coordination-layer-ai-coding-agents) for AI coding agents on one codebase | Platform to build and run AI agents on a company's knowledge and tools |
| Who it's for | Engineering teams running several coding agents | Every team: sales, support, marketing, engineering, data, IT, legal, people |
| The agents | The ones you already run: Claude Code, Codex, Cursor, Hermes, any MCP client | Dust's own agents, built in Dust, on 20+ models (GPT, Claude, Gemini, Mistral, DeepSeek) |
| Knowledge | The code itself: signatures, callers, recent changes, notes left on functions | Company data synced in: Drive, Notion, Confluence, Slack, GitHub, Salesforce, Snowflake and more |
| Repositories | Reads the repo on your machine, live, at every write | GitHub sync of the default branch, read-only; a code sync at most every 8 hours |
| Two agents, one file | A write against code another agent just changed is held, with what changed | Not addressed |
| Renames across files | Reach every agent that calls the old name, on its next step | Not addressed |
| Landing | `git push` rebases onto teammates' commits, reruns the tests and pushes | Not addressed |
| Multi-agent | Several agents, across a team's machines, on one codebase | Agents call other agents as tools (up to 4 deep) and share Pods with people |
| Where it runs | Hooks in your agents, an MCP server, the dashboard, the `cork` command | Web app, Slack, Teams, a browser extension, Zapier, n8n, an API and a CLI |
| MCP | An MCP server, https://mcp.corkcli.com | Uses MCP servers, and serves your workspace at https://dust.tt/mcp |
| Price | Free on one machine; 10 agent hours a month free signed in; $20 per 100 agent hours, pooled | Free seat (500 credits, lifetime); Pro €24 and Max €120 a seat a month, billed yearly; Enterprise on request |

## What is Dust?

Dust calls itself "multiplayer AI for human-agent collaboration": "Dust connects your company knowledge, tools, and teams so you can create, share, and run agents across real workflows." A company connects its data sources (Google Drive, Notion, Confluence, Intercom, GitHub, Snowflake, Zendesk, Slack, Salesforce and more), and anyone can build an agent on them, choose its model and tools, and share it. Agents run where people work: in the web app, in Slack and Teams, in a browser extension, from automation tools such as Zapier and n8n, and through an API and a CLI.

Its multi-agent features are about agents and people working together inside Dust. An agent can call another as a tool, several agents can join one conversation, and Pods are "shared workspaces where humans and agents collaborate around conversations, tasks, and files." Dust is SOC 2 Type II certified and GDPR compliant, hosts in the EU or the US, and says customer data is never used to train models. It was founded in 2023 by Gabriel Hubert and Stanislas Polu and raised a $40m Series B in May 2026, co-led by Abstract and Sequoia.

For developers, Dust offers its CLI, a skill that teaches Claude Code or Codex to talk to Dust agents, and an MCP server that exposes your workspace's agents, conversations and files to an IDE. Its GitHub tool works with issues, pull request reviews and comments, and has no actions for commits, branches or merges.

## What is Cork?

Cork (corkcli.com, by Lithometric) keeps several AI coding agents, and the people running them, from overwriting each other on one codebase, and hands each agent what the others already learned. Hooks in Claude Code, Codex, Cursor and Hermes do the work with no extra model calls, and an MCP server (https://mcp.corkcli.com) serves any other client:

- **Briefing:** the exact signatures, callers and recent changes of the code an agent is about to touch, at session start and with each prompt.
- **Live awareness:** a teammate's rename, removal or new signature reaches every agent that uses it on its next step, as a `Δ` line.
- **[Write gate](https://corkcli.com/blog/write-gate):** a write against code a teammate just changed is held once and told what changed; the retry goes through.
- **[Semantic lint](https://corkcli.com/blog/semantic-lint):** every write is checked against other agents' renames and signature changes, with the exact lines.
- **Landing:** `git push` becomes one step that rebases onto teammates' commits, reruns the tests and pushes.
- **Messages and handoffs:** `cork message` to the repo, one agent or whoever is on a function; `cork handoff` to your other session, which starts on it.

## Where Dust stands out

- **Every team, not just engineering.** Support, sales, marketing and legal build agents in one place, on one set of connected data.
- **Company knowledge.** Dozens of connectors bring documents, tickets, CRM records and warehouse tables into the agents' reach.
- **Agents where people already are.** In Slack and Teams, in the browser, and in automation tools, with no terminal needed.
- **Building agents without code.** An agent builder, skills and schedules, so the people who know a workflow can automate it themselves.
- **Governance for the whole company.** SSO, SCIM, audit logs and EU or US data residency.

## Where Cork wins

- **Agents editing the same code.** Cork sees the code at the level of functions, so it catches a rename in one file that breaks a call in another, and holds the stale write before it lands. Dust has no equivalent.
- **The code as it is now.** Cork reads the repository on the machine at every write. Dust's GitHub connection syncs the default branch, read-only, at most every eight hours, and direct pushes don't trigger a sync.
- **Pushes that land.** With sixteen agents on one repo, rejected pushes fell from 40.6 per run without Cork to 0 with it, because `git push` rebases, tests and pushes in one step.
- **Fewer tokens, measured.** Briefings mean agents search and re-read less: in Cork's benchmarks, 24% fewer tokens with one Claude Code agent and 47% fewer with sixteen at once.
- **The agents you already use.** Cork goes under Claude Code, Codex and Cursor rather than asking your team to move to another agent.

## Can you use both?

Yes, and they fit together, because they don't overlap. Dust holds the company's knowledge and runs the agents that the rest of the business uses. Cork coordinates the coding agents engineers run on the codebase. A Claude Code agent can have both: Dust's MCP server, so it can ask a Dust agent what the support team knows about a bug, and Cork's hooks, so it doesn't overwrite the teammate's agent working in the same module.

## Pricing compared

Dust prices per seat, in euros, excluding VAT, for teams of up to 100 on its Business plan. A Free seat comes with 500 credits for its lifetime. Pro is €24 a seat a month billed yearly (€30 monthly) with 8,000 credits a seat a month, and Max is €120 (€150 monthly) with 40,000. A credit is Dust's unit of AI usage, and unused credits don't roll over. Programmatic use is $0.01 a credit. Enterprise, with unlimited connectors, pooled credits, SCIM and single-tenant deployment, is priced on request.

Cork on one machine with no account is free, with all data on the machine. Signed in, 10 agent hours a month are free per account (30 bonus hours the first month). Beyond that you pay for agent hours, not seats: $20 a month per unit of 100 hours, pooled across the workspace, graduated to $17 (units 11 to 50) and $15 (51 and up), with 15% off a year. An agent hour is time the model is thinking and writing; a build or test run never counts. Enterprise starts from about 100 units.

## Bottom line

Dust is for a company that wants agents across its business, built on its own knowledge, used in Slack and the browser, governed in one place. Cork is for an engineering team whose coding agents are starting to trip over each other on one repository. If you have both problems, you can use both.

## Frequently asked questions

### Is Cork an alternative to Dust?

Not really: they do different jobs. Dust builds and runs agents on a company's knowledge for every team. Cork coordinates the coding agents engineers already run on one codebase, so they don't overwrite each other. Many teams could use both.

### Does Dust coordinate coding agents?

Not as of November 2026. Dust can launch Cursor Cloud Agents on a GitHub repository and follow their pull requests, and its CLI lets Claude Code or Codex query Dust agents. Its documentation has nothing on merge conflicts, renames across files, held writes or landing pushes.

### Can Claude Code use Dust and Cork together?

Yes. Connect Dust's MCP server (https://dust.tt/mcp) so the agent can reach your Dust agents and company knowledge, and install Cork (`npx corkcli`) so its edits are coordinated with the other agents on the repository.

## Related reading

- [Cork vs Amp: one agent in many orbs, or many agents on one codebase](https://corkcli.com/blog/cork-vs-amp): Cork vs Amp: Amp runs its agent in cloud orbs you can share. Cork keeps Claude Code, Codex and Cursor agents from overwriting each other on one repo.
- [Cork vs Claude Code agent teams: one session's crew, or every agent on the repo](https://corkcli.com/blog/cork-vs-claude-code-agent-teams): Cork vs Claude Code subagents, agent teams and cross-session messaging: what each coordinates, where agents collide, and when to use both together.
- [Cork vs AQ: a shared cloud workspace, or a coordination layer for the agents you already run](https://corkcli.com/blog/cork-vs-aq): Cork vs AQ: AQ hosts multiplayer cloud workspaces for coding agents. Cork coordinates the agents you already run so they stop overwriting each other.
- [Cork vs Cursor Projects: a coordinator for Cursor agents, or shared state for every agent](https://corkcli.com/blog/cork-vs-cursor-projects): Cork vs Cursor Projects: a Projects coordinator plans work and directs Cursor's agents with shared context files; Cork keeps any agents on one codebase aware of each other.

[All articles](https://corkcli.com/blog.md) · [Cork in Markdown](https://corkcli.com/index.md) · [llms.txt](https://corkcli.com/llms.txt)

This page in HTML: https://corkcli.com/blog/cork-vs-dust
