[ Deploy ]

Deploy your agent where its model runs.

A b3os.yaml names your agent's services and what they call: the model, packs, routines and sandboxes. A push or b3os apply builds it on the node and switches over once the new version is ready.

CLI
b3os applyb3os app statusb3os app rollback
GitHub app
B3OS Deploy
version: 1name: research-agentinfer:  models:    - id: b3os/default      kind: chatservices:  - name: harness    build:      template: agent      root: agent      lockfiles: [package.json, package-lock.json]      install: npm ci --omit=dev    command: [node, src/main.ts]    readiness:      http: { port: 8080, path: /healthz }    uses: [model/b3os/default, vision]

[01/03] What it does

  1. One file

    b3os.yaml sits in the repository beside your code. The CLI checks it offline and names the line and column of each mistake, down to a routine file that is missing.

    b3os manifest validate

  2. A plan before any change

    It names what is missing, such as a connection you have not stored.

    b3os apply --dry-run

  3. Builds kept as snapshots

    A build installs your apt packages, then runs your install with only the lockfiles present and no secrets. B3OS snapshots the result and reuses it until the lockfiles or the build settings change.

    build: template, apt, install, lockfiles

  4. Undone on failure

    If a step fails or you press Ctrl-C before the cutover, apply deletes what it started and puts each routine back. The running generation keeps serving.

  5. Status, logs and requests

    See each running service's URL, process, last call and lease. Follow its output, and list the calls it made to B3OS, or only the ones that failed.

    b3os app status · logs · requests

  6. Rollback

    Deploy the previous live commit again, without a checkout. A deploy from GitHub rolls back from the console.

    b3os app rollback

  7. In the console

    Applications draws the app as a canvas of its services and what they call, with tabs for its deployments, logs and traces. Wrap a turn in b3.traces.turn and its calls show as one waterfall.

From b3os.yaml to a running app
  1. Plan

    Every item checked against your project. A problem stops here.

  2. Build

    Your install, or the snapshot of the last one.

  3. Start

    A new generation boots from the snapshot with the commit's files.

  4. Ready

    B3OS waits until its readiness probe passes.

  5. Cut over

    The old generation drains and is deleted.

  • Version 1 of b3os.yaml is a draft and can change.
  • A service is a sandbox built from a B3OS template, and counts toward your sandbox limits. Your own Docker images and volumes are not supported yet.
  • A service gets the tracked files of its folder at the commit, so .env and other ignored files never leave your machine.

[02/03] From GitHub

No workflow file and no key in the repository.

  1. Install

    Add B3OS Deploy to the account that owns the repository. It reads files and posts the plan; it cannot push.

  2. Link

    In the console, pick the repository, the branch and the b3os.yaml to deploy.

  3. Push

    B3OS plans the commit, posts the plan as a check, then builds and starts it.

  4. Pull request

    The plan goes on the pull request as a check and a comment. Nothing deploys.

[03/03] Install

CLI
npm install -g @b3os/cli@0.2
TypeScript SDK
npm install @b3os/sdk@0.2
Python SDK
pip install 'b3os>=0.2,<0.3'

Free during the beta.

Invite-only for now.