Tandem SDK

import { LinkCard, CardGrid } from “@astrojs/starlight/components”;

The Tandem engine exposes a full HTTP + SSE API. The official SDKs wrap this API with typed, ergonomic interfaces for TypeScript (Node.js) and Python.

Both SDKs cover the main engine namespaces, including sessions, streaming, memory, skills, channels, MCP, workflows, workflow plans, routines, automations, agent teams, and missions.

Prerequisites

A running Tandem engine and an API token:

tandem-engine serve --api-token $(tandem-engine token generate)
# or via desktop Settings → API Tokens

Packages

Guides

Common event types

Both SDKs expose the same SSE event stream. These are the most common types you will handle:

event.type	Description	Key property
session.response	Text delta from the model	properties.delta
session.tool_call	Tool invocation	properties.tool, properties.args
session.tool_result	Tool result	properties.output
run.complete	Run finished successfully	properties.runID
run.completed	Alternate success event name	properties.runID
session.run.finished	Session-scoped terminal event	properties.status
run.failed	Run failed	properties.error
permission.request	Approval required	properties.permission
question.pending	Structured question from agent	properties.text

Namespace overview

All namespaces exist on both the TypeScript and Python clients.

Namespace	What it covers
sessions	Create, list, message, fork, abort, cancel, diff, revert sessions
permissions	List and reply to permission requests
questions	List/reply/reject AI-generated approval questions
providers	Catalog, config, set API keys and defaults
channels	Telegram, Discord, Slack integration config
mcp	Register, connect, refresh MCP servers and tools
browser	Browser sidecar status, install, and smoke testing only
memory	Global memory: put, search, list, promote, demote, delete, audit
skills	Agent skill packs: list, import, preview, install templates
resources	Key-value resource store (shared agent state)
workflows	Workflow registry, runs, hooks, and event streams
workflowPlans / workflow_plans	Planner chat, preview, and apply flows for automation generation
routines	Scheduled routines: create, run, approve/deny/pause/resume runs
automations	Legacy mission-scoped automations (compatibility path)
automationsV2 / automations_v2	Persistent multi-agent DAG automations with per-agent model policy
bugMonitor / bug_monitor	Incident triage, drafts, approval, and publishing helpers
coder	Coder runs, artifacts, review summaries, and memory candidates
agentTeams	Spawn and manage multi-agent teams
missions	Multi-agent goals and work item tracking

For browser automation itself, use standard engine tools such as browser_open, browser_click, and browser_screenshot through execute_tool(...) or session-based runs with those tools in the allowlist. The browser namespace is for diagnostics and install flows.