Build AI workflows with the people who know the work.
Endeavor connects your engineers with functional teams and domain experts. Experts define the work and what good looks like. Your team builds, integrates and ships it with the Guidelight SDK, the Endeavor API and the coding agents you already use.
- Describe the work and the goal
- Bring real examples
- Score outputs and approve releases
- Connect systems via API or MCP
- Build and compare in code
- Release and run in your apps
One platform for the business and the builders
Start from the business problem
Every workflow is tied to an objective and KPIs the business owns, so you build what matters first and can prove it worked.
Experts in the loop, not in your backlog
Domain experts write test cases and score outputs in the Endeavor workspace. Quality improves without tickets, spreadsheets or prompt edits in your code.
Your stack, your code
Use the Python SDK, the REST API or MCP. Connect the systems you already run and call released workflows from your own applications.
Governed and versioned
Every release is tested, versioned and runs through your governance policies, credentials and logging. Roll forward or back with confidence.
How a technical team builds a workflow on Endeavor
Six steps, from a business goal to a governed workflow in production. Each step shows the tool you use.
Agree on the goal
Pick one business problem with the functional team. Compass turns their process into tasks with an objective, KPIs and human checkpoints.
Set up your project
pip install guidelight, create client credentials for your Endeavor instance, and connect with gdl.connect().
Give your coding agent context
Point Claude Code, Codex or Cursor at the Guidelight docs and the OpenAPI spec at /docs, or connect it to the Endeavor MCP endpoint.
Build and test against expert examples
Generate the integration and experiments, run them against the test cases experts wrote, and let experts score the results.
Release and call it
Promote the version experts approve to staging, then production. Call it from your app with endeavor.run() or the REST endpoint.
Monitor and improve
Track usage, cost and quality. Ship a new version when it beats the current one on the same tests.
You and your coding agent write the code. Endeavor makes it reliable.
Use Endeavor with Claude Code, Codex, Cursor or any coding agent that reads docs or speaks MCP. The agent moves fast. Endeavor makes the result something the business can depend on:
> Read the Guidelight docs and the Endeavor OpenAPI spec at https://<your-endeavor>/docs/openapi.json. Connect our help desk webhook to the "triage-ticket" task in the "support-desk" agent. Run an experiment against the test cases the support team wrote, compare two models, and show me the scores. Then add a call to the staging release in our ticket handler.
Give your agent three things: the Guidelight SDK, the OpenAPI spec at /docs/openapi.json, or the Endeavor MCP endpoint at /mcp.
Support ticket triage and resolution
The problem: tickets land in the wrong queue, first responses are slow, and the best answers live in a few senior agents' heads. Here is how a support team and a technical team build the fix together.
Define the work
Support lead + CompassThe support lead uploads the triage runbook. Compass breaks it into tasks: classify the request, set priority, find the right knowledge-base article, draft a reply, route to the right queue. Each task gets a goal and KPIs: routing accuracy and first-response time.
Set the bar
Support agentsAgents add 50 real past tickets with the correct category and a good reply. These become versioned test cases, and the agents score outputs in the workspace.
Build and compare
DeveloperThe developer connects the help desk and knowledge base through the API or MCP, then runs experiments that compare models and prompts against the agents' test cases.
Release and run
Developer + support leadThe version the support lead approves is released to production and called from the help desk on every new ticket. Agents review every draft before it is sent.
import guidelight as gdl
endeavor = gdl.connect()
def on_new_ticket(ticket): # help desk webhook
triage = endeavor.run(
"support-desk", "triage-ticket",
context={"subject": ticket.subject, "body": ticket.body},
environment="production",
)
ticket.set_queue(triage["queue"]) # route
ticket.set_priority(triage["priority"])
ticket.add_draft(triage["draft_reply"]) # agent reviews and sends
Keep improving
EveryoneAgents keep scoring live outputs. When a new version beats the current one on the same tests, it is released as 1.1.0, with no rewrite.
RFP response generation
Proposal leads define what a winning response looks like and supply past proposals. Developers connect the content library and CRM, then release a task that drafts tailored, compliant first responses for experts to edit.
Pick a use case. Build it together.
24 starting points across eight business functions, each with the systems it connects to. Bring one to your business partners and build it as a team.
Three ways in
Python SDK
Typed models, pagination and authentication for the whole lifecycle: build, evaluate, release and run.
Read the SDK docs →REST API
Any language. Bearer-token auth and an OpenAPI spec, served at /docs on every Endeavor instance.
Getting started →MCP
An MCP endpoint over Streamable HTTP, so MCP-compatible tools and agents can work with Endeavor directly.
See integrations →A few lines to call a released workflow
pip install guidelight
import guidelight as gdl
endeavor = gdl.connect() # uses your client credentials
result = endeavor.run(
"support-desk", "triage-ticket",
context={"subject": "...", "body": "..."},
)
Released workflows also have a stable REST URL: POST /api/{environment}/{agent}/{task}/{version}
Run it where your data needs to live
Use the same SDK, API and workflows wherever you run them, and move to your own cloud or environment when you need to.
Choose your hosting platform
Run the apps and workflows you build on the platform you already use, such as Replit, Render or Vercel. They call Endeavor through the SDK or API.
- Deploy with your existing pipeline
- Call released workflows over HTTPS
- Keep credentials in your platform's secrets
In your cloud
Deploy in your AWS, Azure or Google Cloud account, for example with Amazon Bedrock in your VPC.
- Private Kubernetes namespace
- Egress limited to approved endpoints
- Your secrets stay in your secret manager
On-premises or fully private
Run Endeavor and your models in your own environment, so no data reaches a third-party provider.
- Host open-source models locally
- Your network, your controls
- Built for regulated work
Quick answers
Yes. Point them at the Guidelight SDK docs and the OpenAPI spec, or connect them to the Endeavor MCP endpoint. Endeavor adds the testing, versioning and governance that make their output production-ready.
No. They work in the Endeavor workspace: describing the work, adding test cases and scoring outputs. Developers work in code against the same tasks.
The SDK uses client credentials for your Endeavor instance and handles tokens for you. REST calls use a bearer token.
Each Endeavor instance serves interactive API docs at /docs. The SDK guides are on GitHub, and Guidelight is BSD-3 open source.
With a free Proof of Value: one workflow, built with your business team and your engineers on your own examples.
Pick one business problem. Build it together.
A free Proof of Value brings your business experts and engineers together on one workflow.