Allternit Playground

Overview

Allternit Playground is a session-level debugging surface inside the Allternit platform. It lets you send ad-hoc prompts against a live session, inspect the memory that session has accumulated, toggle which tools the agent can use, and replay the event stream that led to the current state. Use it when you need to reproduce a behavior, verify that a session remembers the right context, or iterate on a system prompt without leaving the platform shell.

How to access it

Allternit Playground runs inside the platform shell:

https://app.allternit.com/shell?view=allternit-playground

Open programmatically from shell code:

open('allternit-playground');

The surface is a singleton view, so only one Allternit Playground tab can be open at a time.

Features

Prompt runner

Send system and user messages to a live session and inspect the raw response

Memory search

Query session memory by keyword and see ranked excerpts with section, line, and score

Tool selector

Attach built-in tools such as web search, bash, text editor, computer use, and MCP registry

Event replay

List every event recorded for a session, including type, timestamp, and full payload

Playground tabs

Tab	Purpose
Prompt	Compose a system instruction and user message, then run them against the target session
Memory	Search the session’s memory store for relevant excerpts
Tools	Choose which tools are available for the playground run
Events	Load and inspect the chronological event log for the session

Session ID

Every tab scopes its work to a single Session ID. Enter the ID in the header input before running prompts, searching memory, or loading events. The Events tab fetches automatically once a session ID is present; the other tabs run on demand.

Prompt tab

The Prompt tab sends a two-message payload to the session run endpoint:

{
  "messages": [
    { "role": "system", "content": "You are a helpful Allternit agent." },
    { "role": "user", "content": "<your message>" }
  ],
  "tools": ["web_search", "bash"]
}

Field	Default	Description
`system`	`You are a helpful Allternit agent.`	System instruction for the run
`user`	`—`	The user message sent to the agent
`tools`	`[]`	Enabled tool IDs for this run

The raw JSON response is rendered in a scrollable output panel. Use Clear output to reset it.

Memory tab

Enter a query and press Search (or hit Enter) to search the session memory. Results show:

Field	Description
`section`	Memory section the match came from
`line`	Line number of the excerpt
`excerpt`	Matching text snippet
`score`	Relevance score

Tools tab

The Tools tab lets you toggle which capabilities the playground run can invoke. Available tool IDs:

Tool ID	Description
`web_search`	Search the web for current information
`bash`	Run shell commands in a sandbox
`text_editor`	Read and edit files
`computer_use`	Control a desktop environment
`mcp_registry`	Use registered MCP tools

Only selected tools are passed to the run endpoint. If no tools are selected, the request omits the tools field.

Events tab

The Events tab loads the full event ledger for the session. Each event displays:

type — the event type
created_at — ISO timestamp
data — full event payload as formatted JSON

Events are fetched automatically when you switch to the tab and a session ID is set.

Endpoints

Method	Path	Description
`GET`	`/api/v1/beta/sessions/{sessionId}/memory/search?q={query}`	Search session memory
`GET`	`/api/v1/beta/sessions/{sessionId}/events/list`	List session events
`POST`	`/api/v1/beta/sessions/{sessionId}/run`	Run a prompt against the session

Request example

curl -X POST https://api.allternit.com/api/v1/beta/sessions/{sessionId}/run \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [
      { "role": "system", "content": "You are a helpful Allternit agent." },
      { "role": "user", "content": "What did we decide about the retry policy?" }
    ],
    "tools": ["memory"]
  }'

Error handling

Allternit Playground surfaces errors as in-app toasts. Common causes:

Symptom	Cause
”Failed to search session memory”	Missing session ID, empty query, or backend error
”Failed to load session events”	Invalid session ID or events service unavailable
”Session ID and user prompt are required”	Prompt tab submitted without both fields filled
”Run failed”	The `/run` endpoint returned a non-2xx response

Toasts auto-dismiss after five seconds and can be closed manually.

Key concepts

Session-first debugging — every action is scoped to a single session ID, so you can inspect the exact context an agent is working with.
Ad-hoc runs — send one-off prompts without creating or saving an agent configuration.
Tool attachment — toggle tools per run to reproduce issues caused by specific capabilities.
Memory inspection — verify what the session remembers and how relevant memories are ranked.
Event replay — trace the sequence of events that produced the current session state.

Get Started

Core Concepts

Gizzi Code CLI

Tools & ACI

Security & Enterprise

Admin API

BYOC Runtime

Surfaces

CI CD

Guides

Cookbooks

Release Notes

Overview

How to access it

Features

Prompt runner

Memory search

Tool selector

Event replay

Playground tabs

Session ID

Prompt tab

Memory tab

Tools tab

Events tab

Endpoints

Request example

Error handling

Key concepts

​Overview

​How to access it

​Features

Prompt runner

Memory search

Tool selector

Event replay

​Playground tabs

​Session ID

​Prompt tab

​Memory tab

​Tools tab

​Events tab

​Endpoints

​Request example

​Error handling

​Key concepts

​Related links

Overview

How to access it

Features

Playground tabs

Session ID

Prompt tab

Memory tab

Tools tab

Events tab

Endpoints

Request example

Error handling

Key concepts

Related links