Define your agent.
We do the rest.
Install the CLI, scaffold an agent, and push it to the registry. Astropods handles containers, secrets, scaling, and observability from the first run.
The spec is the deploy form.
Declare an input, a model, or a provider in astropods.yml and it becomes a field the moment the blueprint deploys.
astropods.ymlBring the framework you already use.
An adapter wraps your agent where it stands. The code keeps its own shape, the spec adds the runtime around it, and nothing gets rebuilt to ship.
Build the agent. The rest is already running.
No registry to build, no tracing to bolt on, and no eval harness to write. These are built in for every blueprint.
Push an agent blueprint once and the whole team can find it, run it, and build on it. Each deploy is one click and its own agent, with its own settings and its own spend.
All the infrastructure you need, already there.
No store to provision, no model key to rotate, no token in your repo. The blueprint arrives and the runtime is already there.
One line in the spec and the agent gets a managed key, an OpenAI-compatible URL, and the models you picked at deploy.
Name a vector store, a cache, or a database in the spec and it exists when the agent deploys. Other agents can read the same one.
Keep every token in the account vault and name it in the spec. The value is handed to the container at deploy and never lands in your repo.
Turn messaging on in the spec and a sidecar puts the agent on Slack and web chat. Every platform arrives as the same event, so there is no integration to write for each one.
Explore agent blueprints
Use a blueprint that already works, add your own keys, and deploy it under your account.
Browse all blueprints
Watches every channel it is invited to and works out which discussions a team lead needs. Urgent ones arrive as a DM, the rest in a twice-daily digest.
Reviews pull requests for correctness, security, and quality, with one-click commenting and a fix prompt for your coding agent.
Writes release notes from the Jira issues and GitHub pull requests that went into the release.
A personal knowledge base. Save anything and find it again by asking for it in plain language.
A Slack-first blog pipeline. Drop a topic and it researches, drafts, copy-edits, and stages the post for review.
Surfaces issue trends, categories, competitor mentions, and solutions across a repository, with no query to write.
Connects to Greenhouse and ranks your pipeline against any open req, with multi-feature scoring and a second-pass critic review.
Or bring your own agent
Questions builders ask
An account and Docker. Install the CLI, run ast project create, and set the credentials it asks for. Astropods builds the image and runs it. There is no cluster to stand up first.
One file, astropods.yml, describes the whole agent: where it builds from, the models it calls, what it can reach, and what somebody has to fill in. Push it and you have a blueprint, a versioned snapshot in the registry. Deploy that and you have a live agent, which anyone in the organization can run or start from without opening the YAML.
Yes. There are adapters for Google ADK, OpenAI, CrewAI, LangChain, LangGraph, the Vercel AI SDK, Mastra, and more. If you already have a Dockerfile, point the spec at it and Astropods takes the runtime around your code.
Bring them over. There is no rewrite: the agent keeps its code, its framework, and the Dockerfile you already ship. One file beside it is the whole port, and it can run next to what you have today until you are ready to move the traffic.
Yes. ast project start runs the agent on your machine against the same spec the deploy reads, so what you test is what you push.
Yes, and it is as easy as adding one line to the spec. Set frontend: true and the agent gets its own HTTPS hostname, with sign-in already at the front door. Anything that speaks HTTP will do.
In the account vault. The spec refers to them by name, as KEY=@SECRET_NAME, so no value is in your repo or your image. The agent gets them at run time.
Yes. Astropods runs managed, in your VPC, or fully isolated for regulated environments. The spec and the CLI do not change.