Skip to documentation
Documentation
Documentation
OverviewInstallationQuickstartYour first dayYou were sent a linkWorking with a customerConfigurationTroubleshootingFAQGlossary

Usage

Terminal interfaceCLIStatuslineDashboardAnalytics

Collaborate

DocumentsLessonsSession sharingPicking up tomorrow and handing work overFrom your phoneSession linksSession messagingAsksAsks privacy fact sheetFeedbackEmail from crewkitBlueprintsAgent identitiesMCP server

Configure

AgentsSkillsSkill and agent security scanningRulesCommandsInheritancePlaybooksExperimentsHooksCapture modesSensitive sessionsData residencyTeamSettings mapIntegrations

Workspaces

WorkspacesSet up a workspaceFix a ticket end to endShip a feature with approvalHand off work and set agent settings

API Reference

API OverviewAuthenticationSessions APIResources APIOrganizations APIProjects APIPlaybooks APIExperiments API
Need a hand?Find an answer in troubleshooting.
Loading docs…
CREWKIT / DOCUMENTATION

Documents

Project memory — upload docs, import from Drive, capture Slack and email, and search everything your project knows.

Documents are your project's memory: PRDs, meeting transcripts, contracts, design docs, captured Slack messages, and past conversations. Everything becomes searchable, and the most relevant pieces are injected automatically when a session starts — so decisions stop evaporating when the conversation ends.

The API calls documents artifacts: its paths and fields use that word. See the API reference.


Upload documents

Go to Projects → your project → Documents in the dashboard. Upload with the button or drag and drop files onto the page.

Accepted formats: PDF, Markdown, plain text, and Word (.doc/.docx).


Import from Google Drive

From the same Documents page, connect Google Drive and import files directly. The connection is authorized per organization; imported files become project documents like any upload.


Capture from Slack and email

Slack messages and email directed at a project's active agent identity are captured as documents. Because inbound content is untrusted, crewkit does not add it to a session's context on its own until a person confirms it in the Inbox. It can be found by search, by members and by agents, as soon as it arrives. Approve trusted items to make them part of project memory; delete the rest.

Project email addresses are listed on the project overview, where each one can also be downloaded as a contact card. Only active identity addresses accept and capture inbound email; paused and archived addresses do not. Inbound email receives no reply from the agent; its purpose is to create a reviewable document without exposing unconfirmed content to agent context.


Search project memory

The Documents page has a search box that searches everything the project knows: uploaded docs, Drive imports, captured messages, and indexed past conversations. Ask in plain language. Search is semantic: crewkit sends your query to its embedding provider, Voyage AI, and ranks results by meaning, not just keyword match. crewkit falls back to keyword search when the query cannot be sent, because it came from an untrusted source or your organization's data settings bar it, or when semantic matching is unavailable.

Mid-session, agents can run the same search through the crewkit MCP server, which helps when the answer isn't in the codebase (decisions, requirements, prior discussions).

For an organization that keeps its data on its own data plane, project search is not available.


Automatic context injection

When you run crewkit code, the CLI fetches the most relevant documents for your project and injects them into the session context. In headless mode (), your prompt is the search query. An interactive session searches with the fixed query "general project context". Results are added from most to least relevant, and any result that would push the block past about 8,000 characters is left out. After launch, each prompt you type is sent as a search query too, unless the session is , was started with , or the capture mode is .

Analytics

Track session KPIs, costs, and agent performance across your team.

Lessons

Short rules your team learned from its own repeated corrections, review comments and confirmed discussions, added to later sessions on the same project. What is collected and why, consent, residency and erasure.

On this page

Upload documentsImport from Google DriveCapture from Slack and emailSearch project memoryAutomatic context injectionFind prior workNext steps
--prompt
--sensitive
--no-artifacts
minimal

A --sensitive session makes no context request at launch, and neither does a launch for an organization that keeps its data on its own data plane, when the CLI knows that at launch. When crewkit did not answer the launch, a request carrying your prompt can reach crewkit, which refuses it. Later prompts are searched on your data plane, for Lessons only.

crewkit code                  # Context injected automatically
crewkit code --no-artifacts   # Skip document context injection

If the API is unreachable, injection degrades silently — your session still launches.


Find prior work

Every analyzed conversation is indexed into project memory too. Before starting non-trivial work, ask "has anyone touched this before?" and find the actual conversation that did the work — via the Documents search or the MCP find_prior_work tool mid-session.

Sensitive and private conversations stay privacy-gated: they are excluded from cross-conversation discovery.

What a conversation contributes depends on your organization's capture mode. For what crewkit captures from each session, and what it stores and indexes, see Capture modes.


Next steps

  • MCP server
  • Agent identities
  • Blueprints