Overview
The crewship.toml file configures how your crew is built and deployed. Place it in your project root.
Crewship supports two formats:
- Single deployment — one
[deployment] section (default, created by crewship init)
- Multi-deployment — multiple
[deployments.<name>] sections for monorepos with several agents
Run crewship init to auto-generate this file with detected settings from your project. Single Deployment (default)
CrewAI
LangGraph
LangGraph.js
[deployment]
framework = "crewai"
entrypoint = "src.my_crew.crew:MyCrew"
profile = "slim"
python = "3.11"
[build]
exclude = ["tests"]
[deployment]
framework = "langgraph"
entrypoint = "src.my_agent.graph:graph"
profile = "slim"
python = "3.11"
[build]
exclude = ["tests"]
[deployment]
framework = "langgraph-js"
entrypoint = "./src/graph.ts:graph"
profile = "slim"
[build]
exclude = ["tests"]
Multi-Deployment
For monorepos with multiple agents sharing the same codebase, use named [deployments.<name>] sections:
[build]
exclude = ["tests"]
[deployments.research-agent]
framework = "crewai"
entrypoint = "research_crew.crew:ResearchCrew"
profile = "slim"
python = "3.11"
[deployments.writer-agent]
framework = "crewai"
entrypoint = "writer_crew.crew:WriterCrew"
research-agent).
[apis] and [chat] can be set at the top level as global defaults, and overridden per deployment:
# Global defaults
[chat]
input_key = "query"
[deployments.support-agent]
framework = "crewai"
entrypoint = "support_crew.crew:SupportCrew"
[deployments.research-agent]
framework = "crewai"
entrypoint = "research_crew.crew:ResearchCrew"
# Override chat config for this deployment only
[deployments.research-agent.chat]
input_key = "topic"
[deployment] and [deployments.*] are mutually exclusive. Using both in the same file will produce an error.
Selecting a deployment
Use --name / -n on any command to target a specific deployment:
crewship deploy --name research-agent
crewship invoke --name writer-agent --input '{"topic": "AI"}'
crewship env set --name research-agent OPENAI_API_KEY=sk-...
--name is omitted:
- One named deployment — auto-selected
- Multiple named deployments — interactive prompt (or error in CI)
- Single
[deployment] format — used directly (no --name needed)
Deployment IDs
After the first deploy, deployment_id is saved into each section automatically:
[deployments.research-agent]
framework = "crewai"
entrypoint = "research_crew.crew:ResearchCrew"
deployment_id = "dep_xxx1" # auto-populated
[deployments.writer-agent]
framework = "crewai"
entrypoint = "writer_crew.crew:WriterCrew"
deployment_id = "dep_xxx2" # auto-populated
Full Example
[deployment]
framework = "crewai"
entrypoint = "src.research_crew.crew:ResearchCrew"
python = "3.11"
profile = "slim"
[build]
exclude = ["tests"]
[build.install]
packages = ["ffmpeg", "imagemagick"]
[apis]
enabled = ["thread", "run"]
[chat]
input_key = "topic"
[runtime]
timeout = 600
memory = 512
[metadata]
description = "A crew that researches topics and writes reports"
tags = ["research", "writing"]
Configuration Reference
[deployment] / [deployments.<name>]
| Field | Type | Required | Description |
|---|
framework | string | ✅ | Framework to use: crewai, langgraph, or langgraph-js |
entrypoint | string | ✅ | Entry point for your agent (format varies by framework) |
python | string | "3.11" | Python version (3.10, 3.11, 3.12) — Python frameworks only |
profile | string | "slim" | Base image profile |
dockerfile | string | — | Custom Dockerfile path |
deployment_id | string | — | Auto-populated after first deploy |
Entrypoint Format
The entrypoint format depends on your framework:
CrewAI and LangGraph (Python) — Python module path:
module.path:ClassName_or_variable
# CrewAI: points to a class decorated with @CrewBase
entrypoint = "src.my_crew.crew:MyCrew"
# LangGraph: points to a compiled graph variable
entrypoint = "src.my_agent.graph:graph"
./path/to/file.ts:exportName
# LangGraph.js: points to an exported compiled graph
entrypoint = "./src/graph.ts:graph"
Run crewship init to auto-detect your framework and entrypoint. For LangGraph projects, place a langgraph.json in your project root to enable auto-detection.
Profile Options
| Profile | Description | Frameworks | Use case |
|---|
slim | Minimal base environment | All | Most agents |
browser | Includes Playwright + Chromium | crewai, langgraph | Web scraping, screenshots |
[build]
Build configuration options.
| Field | Type | Default | Description |
|---|
exclude | array of string | [] | Paths to exclude from build |
[build.install]
Additional system packages to install.
[build.install]
packages = ["ffmpeg", "poppler-utils", "tesseract-ocr"]
Only packages available in the base image’s package manager (apt) are supported.
[runtime]
Runtime configuration.
| Field | Type | Default | Description |
|---|
timeout | integer | 300 | Max execution time in seconds |
memory | integer | 256 | Memory limit in MB |
[runtime]
timeout = 900 # 15 minutes
memory = 1024 # 1 GB
[apis]
Controls which APIs are enabled for the deployment. This affects both direct API access and which interaction modes are available in Slack.
| Field | Type | Default | Description |
|---|
enabled | array of string | — | Which APIs to expose: "thread", "run" |
null), all APIs are enabled by default.
[apis]
enabled = ["thread", "run"] # Both APIs enabled (same as default)
[apis]
enabled = ["run"] # Only stateless runs — no thread/conversation support
[apis]
enabled = ["thread"] # Only threaded conversations — no standalone runs
In Slack, the enabled APIs determine which interaction modes work:
- Thread API —
@mention conversations use threads for multi-turn context
- Run API —
/crewship run slash commands use single stateless runs
- If both are enabled,
@mention defaults to thread mode
Not sure which to use? Read Runs vs Threads: When to Use Which. [chat]
Configures how chat messages (from Slack or other integrations) map to your crew’s input and output.
| Field | Type | Default | Description |
|---|
input_key | string | "input" | The parameter name that receives the chat message |
output_key | string | — | Field in the run output containing response messages |
[chat]
input_key = "topic"
{ "topic": "Tell me about quantum computing" }
output_key is set, the platform extracts that field from the run result to display as the response:
[chat]
input_key = "query"
output_key = "messages"
If your crew already accepts an input parameter, you can omit [chat] entirely — it defaults to input_key = "input".
| Field | Type | Description |
|---|
description | string | Human-readable description |
tags | array | Tags for filtering in Console |
[metadata]
description = "Researches topics and generates blog posts"
tags = ["content", "blog", "research"]
Validation
The CLI validates your crewship.toml on deploy:
Common validation errors:
❌ Error: Invalid crewship.toml
- name: must be lowercase with hyphens only
- deployment.entrypoint: required field missing
Environment-specific Config
For different environments, use separate projects:
# Production
crewship deploy --project my-crew
# Staging
crewship deploy --project my-crew-staging
crewship env set OPENAI_API_KEY=sk-prod-... --project my-crew
crewship env set OPENAI_API_KEY=sk-test-... --project my-crew-staging
Example Configurations
Basic CrewAI
[deployment]
framework = "crewai"
entrypoint = "src.simple_crew.crew:SimpleCrew"
Basic LangGraph
[deployment]
framework = "langgraph"
entrypoint = "src.my_agent.graph:graph"
python = "3.11"
Basic LangGraph.js
[deployment]
framework = "langgraph-js"
entrypoint = "./src/graph.ts:graph"
Web Scraping Crew
[deployment]
framework = "crewai"
entrypoint = "src.scraper.crew:ScraperCrew"
profile = "browser"
python = "3.11"
[runtime]
timeout = 600
memory = 1024
Document Processing
[deployment]
framework = "crewai"
entrypoint = "src.processor.crew:ProcessorCrew"
python = "3.11"
[build.install]
packages = ["poppler-utils", "tesseract-ocr"]
[runtime]
timeout = 900
memory = 2048
Slack Chatbot
[deployment]
framework = "crewai"
entrypoint = "src.support_bot.flows.chat_flow:ChatFlow"
python = "3.11"
[apis]
enabled = ["thread"]
[chat]
input_key = "query"
output_key = "messages"
Multi-Agent Monorepo (mixed frameworks)
[build]
exclude = ["tests", "notebooks"]
[deployments.research-agent]
framework = "crewai"
entrypoint = "agents.research.crew:ResearchCrew"
profile = "browser"
python = "3.11"
[deployments.writer-agent]
framework = "langgraph"
entrypoint = "agents.writer.graph:graph"
python = "3.11"
[deployments.frontend-agent]
framework = "langgraph-js"
entrypoint = "./agents/frontend/graph.ts:graph"
crewship deploy --name research-agent
crewship deploy --name writer-agent
