Skip to main content
Every superserve run creates a session - an isolated sandbox running your agent, with its own filesystem and process state. Sessions persist after the conversation ends. Come back hours or days later and resume where you left off.

​
Resume a conversation

$ superserve sessions list my-agent

ID             TITLE                              STATUS   MSGS   LAST ACTIVE
--------------------------------------------------------------------------------
a1b2c3d4e5f6   Review auth module error handling   idle     4      2 minutes ago
f6e5d4c3b2a1   Refactor database migrations        idle     12     1 hour ago

$ superserve sessions resume a1b2

Resumed: "Review auth module error handling" (my-agent)

You> Did you apply both fixes?

Yes — both issues in auth/middleware.py are fixed. Line 47 now returns
a generic 500 with the original exception logged via structlog. Line 82
uses `except Exception:` instead of bare `except:`.

Completed in 2.1s
The agent remembers the full conversation. Files it created or modified are still in the workspace.

​
How persistence works

Sessions use two layers of persistence:
  1. Workspace filesystem - Superserve persists the sandbox’s /workspace directory. Files your agent reads, writes, or modifies survive across turns and restarts.
  2. Conversation memory - Make sure your agent supports multi-turn conversations so it remembers context across messages. Most frameworks handle this automatically when configured correctly. 
# Claude Agent SDK — continue_conversation enables conversation memory
options = ClaudeAgentOptions(..., continue_conversation=True)

​
Session lifecycle

StatusMeaningResumable?
activeProcessing a message right nowNo
idleWaiting for the next messageYes
completedEnded by userYes - fresh sandbox, same workspace
failedCrashed or timed outYes - fresh sandbox, same workspace
# View details
superserve sessions get a1b2

# End a session (workspace is preserved)
superserve sessions end a1b2

​
Sessions in the SDK

You can also manage sessions programmatically with the TypeScript SDK: 
import Superserve from "superserve"

const client = new Superserve({ apiKey: "your-api-key" })
const session = await client.createSession("my-agent")

const r1 = await session.run("What files handle authentication?")
const r2 = await session.run("Add rate limiting to the login endpoint")

await session.end()
The session maintains context between run() and stream() calls - the agent remembers previous messages and the workspace retains all files.