Open-source, secure environment with real-world tools for enterprise-grade agents.
Python agentaiai-agentai-agentscode-interpretercopilotdevelopmentdevtoolsgptgpt-4javascriptllmnextjsopenaipythonreactsoftwaretypescript
Add overview for how smol dev deployment should work
Tomas Valenta8ff9ef58418f
diff --git a/SMOL_DEPLOYMENT.md b/SMOL_DEPLOYMENT.mdnew file mode 100644index 000000000..a3d3c34ce--- /dev/null+++ b/SMOL_DEPLOYMENT.md@@ -0,0 +1,16 @@+# How the smol agent specific deployment works+- Deploy button in repo opens page `/agent/smol-developer`+- On this page user signs in with GitHub+ - We need to fix GH Apps vs OAuth Apps for the sign in - we need GH App access token to retrieve repos that were given permissions by the user+- User selects repo to deploy+ - They can also modify permissions for our GH App if we don't have access to repo they want to use+- User adds initial prompt that specifies what should smol dev do in the first iteration+- User specifies their OpenAI key and clicks on "Deploy"+- Then we create a PR throught the GH API+- We then create agent deployment for the specific repo+ - The necessary GH data are in the `auth` field in the deployments table+ - The OpenAI key is included in the `config` field+ - We also add informations about the PR that was created so we can update the branch connected to the deployment via git pushes in the agent dev environemnt+- During each agent run we authenticate with GH App token, pull the repo, make the changes necessary and then push the changes to the branch connected to the deployment+- On each comment to the PR we trigger a webhook that compiles the prompt from the PR description and comments and then triggers a new run of the smol agent+ - If the smol dev agent is already processing previous PR change we cancel that run and then start a new one with the latest changes