System Manager Tools - Genesis Computing

Overview

Tool Name

system_manager_tools

Purpose

The system_manager_tools group centralizes platform administration plus prior-conversation search, thread export, agent and mission export/import workflows, thread cost analysis, and thread memory recovery. Generate chat titles automatically, search prior conversations, read thread history, analyze cache-aware thread costs, export thread packages and reusable ZIP bundles, re-import agent and mission bundles when needed, compress long-running threads, recover compressed artifacts, check overall health, toggle maintenance mode, adjust configuration, perform safe cleanups, and restart components when required. Most administrative actions are routed through a single entrypoint for predictable governance.

Key Features & Functions

Auto-Title Conversations

Generate concise chat titles that reflect conversation context.

Search Prior Conversations

Search prior threads across one or more bots and read clean thread history by default, or include tool requests and outputs when execution details matter.

Export Threads, Agents, and Missions

Save thread, agent, and mission packages into bot_git repos with download-ready links for handoff or backup, then re-import agent and mission ZIP bundles later.

Thread Cost Analysis

Calculate cache-aware thread token and cost breakdowns, including optional child-thread rollups and pricing overrides.

Thread Compression & Recovery

Compress long-running threads and recover summarized tool results or logically hidden messages without rerunning prior tools.

Health & Status

Retrieve real-time system status with optional connectivity checks.

Maintenance Control

Enable or disable maintenance mode with messages and timing.

Configuration Management

Get, set, update, or reset system configuration keys safely.

Cleanup Operations

Remove temporary files, old logs, or expired sessions with dry-run support.

Targeted Restarts

Restart the platform or specific components with explicit confirmation.

Audit & Logging

Keep an administrative trail of system actions and outcomes.

Organization Health Audit

Start a background, read-only review of the deployment’s skills, blueprints, and live agent activations through the GenesisAuditor.

Check SYSTEM_STATUS before toggling maintenance or performing a restart. It helps confirm health, dependencies, and connectivity.

Input Parameters for Each Function

Most administrative operations use manage_system with an action value. Prior-conversation search uses dedicated find_threads and read_thread functions. Cost analysis uses the dedicated thread_cost function. Export workflows use the dedicated export_thread, export_agent, and export_mission functions. Agent and mission re-import use the dedicated import_agent and import_mission functions. Thread memory management (compression, recovery, status) is handled by manage_thread_memory in context_tools (always available to agents). Poll-and-check pacing uses wait in context_tools (always available to agents). The internal compress_thread function remains available for support workflows but is not exposed as an agent tool. audit_organization_health starts the hidden GenesisAuditor worker. See GenesisAuditor tools for its focus parameter and the read-only audit tools available to the spawned agent.

`AUTO_TITLE_CHAT`

Parameters

Name	Definition	Format
conversation_id	Conversation to generate a title for.	String
conversation_content	Optional summary to guide title quality.	String
title_length	Max title length.	Integer

When OpenAI is the active provider, auto-titling uses the saved fast_model assignment from Config → AI Providers → Change Models (falling back to supervisor or default_chat when fast is unset). The call routes through the LiteLLM engine with per-call OpenAI credentials from AI Providers — not a raw OpenAI client. Genesis applies the assignment on the next title generation without a server restart. For Databricks, auto-titling likewise uses fast_model, falling back to supervisor and then default_chat when the fast assignment is unset. See Databricks provider setup for the model defaults and authentication details.

`SYSTEM_STATUS`

Parameters

Name	Definition	Format
include_details	Include component level details.	Boolean
check_connectivity	Run connectivity probes where available.	Boolean

`MAINTENANCE_MODE`

Parameters

Name	Definition	Format
enable	Turn maintenance mode on or off.	Boolean
maintenance_message	Optional banner message for users.	String
estimated_duration	Estimated duration in minutes.	Integer

`SYSTEM_CONFIG`

Parameters

Name	Definition	Format
config_operation	One of `GET`, `SET`, `UPDATE`, `RESET`.	String
config_key	Configuration key to read or modify.	String
config_value	New value when setting or updating.	String

`CLEANUP_TASKS`

Parameters

Name	Definition	Format
cleanup_type	One of `temporary_files`, `old_logs`, `expired_sessions`, `all`.	String
retention_days	Keep newer items for this many days.	Integer
dry_run	Preview what would be removed without deleting.	Boolean

`SYSTEM_RESTART`

Parameters

Name	Definition	Format
component	Optional component name to restart.	String
force_restart	Skip graceful shutdown if true.	Boolean
confirmation	Must be true to proceed.	Boolean

Restarts interrupt running jobs and sessions. Use SYSTEM_STATUS first, schedule a window, then set confirmation = true only when ready.

`find_threads`

Search prior conversations across one or more bots using hybrid semantic plus lexical matching. Results are ranked by relevance and can be returned as formatted text or structured data. Parameters

Name	Definition	Format
query	Search query. Required. Supports keywords, phrases, or natural language.	String
limit	Maximum threads to return. Default: `5`. Min: `1`. Max: `50`. If Genesis cannot coerce the supplied value to an integer, it falls back to `5`.	Integer
include_summary	Generate LLM summaries for matching threads. Default: `false`. Genesis summarizes at most the top `5` results.	Boolean
include_hidden	Include hidden or background-job threads in results. Default: `false`.	Boolean
bot_ids	Optional list of bot IDs to search. Defaults to the current bot from context when omitted. Use the exact `bot_id` values Genesis returns for agents, such as the values from `GET /api/bots`.	Array of strings
response_format	Response shape: `formatted` or `structured`. Default: `formatted`.	String
stages	Optional staged search sequence. Allowed values: `tokens`, `semantic`, `phrase`.	Array of strings
tool_filter	Optional tool or function name filter for structured tool-call search.	String
tool_action	Optional action filter extracted from tool parameters.	String
param_filter	Optional exact-match filter over indexed tool-call parameters. Nested keys such as `spec.backend` are supported.	Object
result_filter	Optional structured-result filter. Currently supports only `status`, for example `{"status": "FAILED"}`.	Object

bot_id, thread_id, and user_id are injected from the current tool context. A caller identity is required: if user_id does not reach find_threads, it returns an access-denied error with no results rather than searching without an identity filter. find_threads excludes the current thread automatically and still enforces normal visibility and permission checks on returned threads. When you pass bot_ids, use the exact, case-sensitive values from GET /api/bots or _list_all_bots. Cross-bot search enumerates active agents registered in the database, not storage directories without a matching bot row. Structured results include bot_id for each thread; pass that value to read_thread as thread_bot_id when opening a hit owned by a different bot.

Example

find_threads(query="database migration")

find_threads(query="ETL pipeline", limit=10)

find_threads(
    query="UNRESOLVED_COLUMN",
    tool_filter="databricks_action",
    response_format="structured",
)

`read_thread`

Read the content of a specific thread after find_threads or search_threads. Threads are stored per bot; when the hit belongs to another bot, pass that bot’s ID in thread_bot_id using the bot_id from the search result. By default Genesis returns a clean user/assistant transcript and omits tool banners, tool-call payloads, and internal metadata. Set include_tool_calls=true when you need prior SQL, tool outputs, artifact links, reports, or other execution details. Parameters

Name	Definition	Format
thread_id	Thread ID to read. Required.	String
last_n_messages	Optional tail limit for returned user/assistant messages. Valid integers are clamped to `1`-`100`. When omitted, Genesis returns all filtered messages. If the supplied value cannot be coerced to an integer, Genesis falls back to the last `20` messages.	Integer
include_tool_calls	Optional boolean that includes assistant tool requests and tool outputs. Default: `false`. Use this when you need prior SQL, artifact links, reports, or other execution details from the thread.	Boolean
thread_bot_id	Bot that owns the thread when it is not the current bot. Use the `bot_id` reported for the thread by `find_threads` or `search_threads`. Default: the current bot from context.	String

bot_id and user_id are injected from the current tool context. Naming another bot in thread_bot_id does not widen access — Genesis still enforces normal thread visibility on every read. When a thread cannot be read under the requested bot, Genesis returns an error that suggests trying thread_bot_id if another bot may own the thread. If the tool trims the output, the response includes a summary line showing how many messages were returned out of the total message count for the selected mode.

With include_tool_calls=true, Genesis returns unredacted tool calls only to the thread owner or explicitly invited users. Other permitted readers receive redacted tool content and tool_calls_redacted=true in the response.

Example

read_thread(thread_id="chat_abc123")

read_thread(thread_id="chat_abc123", last_n_messages=20)

read_thread(thread_id="chat_abc123", include_tool_calls=True)

read_thread(thread_id="chat_abc123", thread_bot_id="Adam")

`export_thread`

Export the current thread or a specified thread into a bot_git repo as Markdown, PDF, or a ZIP package for backup, import, or support workflows. Support-oriented ZIP exports can bundle bug_report.md, server.log, and any selected file attachments.

When format="pdf", Genesis renders each message header with an ISO-8601 timestamp including timezone offset when timestamp metadata is available. ZIP exports use the same timestamped transcript for thread.pdf, while thread.md remains the clean Markdown transcript.

For PDF availability in Databricks App deployments and ZIP fallback behavior, see Chat and Threads.

In container-based deployments, Genesis prepares the mounted bot_git workspace on startup so export bundles can be written to shared repos even when /app/bot_git is backed by a reused persistent volume.

Parameters

Name	Definition	Format
repo_id	Target `bot_git` repo where the export is saved. Required.	String
source_thread_id	Thread ID to export. Defaults to the current thread from context.	String
format	Export format: `md`, `pdf`, or `zip`. Default: `zip`.	String
filename	Optional custom filename. Auto-generated when omitted.	String
relative_path	Directory inside the repo. Default: `exports`.	String
include_files	Include file attachments in ZIP exports. Default: `true`.	Boolean
include_tool_calls	Include `tool_calls` arrays and tool messages in JSON data (`thread.json`). `thread.md` is always clean Markdown. Default: `true`.	Boolean
bug_report	Markdown bug report written to `bug_report.md` in ZIP exports only.	String
include_server_log	Include the server-side `server.log` file in ZIP exports. Defaults to `true` when `bug_report` is provided and `false` otherwise.	Boolean

bot_id, thread_id, and user_id are injected from the current tool context. user_id is used to enforce thread access, so callers can export only threads they are allowed to access.

When bug_report is provided, Genesis includes server.log in the ZIP by default. Workspace admins can disable server-log inclusion in support bundles with support.include_server_log_in_bug_reports (default: true).

ZIP exports also include scrub_manifest.json. Ordinary export_thread calls keep original thread content intact and record manifest status: disabled so the bundle remains suitable for import. Support-bundle workflows opt into scrubbing separately, which redacts supported text artifacts such as thread.md, thread.json, message_log.csv, bug_report.md, and server.log, while binary attachments remain passthrough.

Export metadata includes message_log_row_count, the number of source rows used to render message_log.csv. When a legacy pre-rendered CSV is reused and its source-row count is unavailable, the value is null; Genesis does not guess a count from rendered CSV lines.

Export metadata also includes message_log_source; see the Thread Export API reference for its values and how to interpret synthesized rows.

Example

export_thread(repo_id="uploads", format="zip", relative_path="exports/")

export_thread(
    repo_id="support",
    source_thread_id="chat_123",
    format="zip",
    bug_report="## Issue\nExport prepared for support review."
)

Successful export_thread, export_agent, and export_mission responses return root-relative Genesis links in download_url; export_thread also returns view_url. Expect /api/git/... in standalone and local deployments, and /git/... in SPCS.

`export_agent`

Export an agent configuration as a ZIP package into a bot_git repo so it can be downloaded or re-imported later. Parameters

Name	Definition	Format
target_bot	Bot ID to export. Defaults to the current bot from context.	String
target_bot_id	Deprecated alias for `target_bot`. Falls back to the current bot from context when used.	String
repo_id	Target `bot_git` repo where the export is saved. Required.	String
relative_path	Directory inside the repo. Optional.	String

Example

export_agent(repo_id="uploads", target_bot="sales_assistant")

`export_mission`

Export a complete mission as a ZIP package into a bot_git repo, including project metadata, thread and replay data, and optional repo, query-result, and agent content. Parameters

Name	Definition	Format
project_id	Mission ID to export. Required.	String
repo_id	Target `bot_git` repo where the export is saved. Required.	String
relative_path	Directory inside the repo. Optional.	String
include_repos	Include Git repositories in the export. Default: `true`.	Boolean
include_query_results	Include query-result caches in the export. Default: `true`.	Boolean
include_agent	Include agent configuration in the export. Default: `false`.	Boolean

Example

export_mission(
    project_id="mission_123",
    repo_id="uploads",
    include_agent=True
)

export_agent and export_mission write binary ZIP files to the target bot_git repo. Storage-only repos such as uploads persist the ZIP on disk without creating a git commit; git-backed repos also stage and commit the export. If the ZIP write fails, the tool returns success: false with an error describing the target path. The API export reference describes how these binary tool-generated bundles handle repository directory content.

`import_agent`

Import an agent ZIP package that was previously exported with export_agent. Genesis assigns a new bot_id for the imported agent so it does not collide with the original. Parameters

Name	Definition	Format
repo_id	Source repo containing the agent ZIP file. Required.	String
file_path	Path to the agent ZIP file within the repo. Required.	String
new_name	Custom name for the imported agent. Optional.	String
handle_conflict	How to handle name conflicts: `error`, `rename`, or `replace`. Default: `rename`.	String

bot_id, thread_id, and user_id are injected from the current tool context. For managed remote repos, Genesis reads the ZIP from the current remote file state instead of a shared local clone. If managed-remote status cannot be verified, the tool returns error_code="MANAGED_REMOTE_STATUS_UNKNOWN".

Example

import_agent(
    repo_id="uploads",
    file_path="exports/sales_assistant.zip",
    new_name="sales_assistant_copy"
)

`import_mission`

Import a mission ZIP package that was previously exported with export_mission. Genesis assigns new IDs for imported entities so the imported mission can coexist safely with the source mission. Parameters

Name	Definition	Format
repo_id	Source repo containing the mission ZIP file. Required.	String
file_path	Path to the mission ZIP file within the repo. Required.	String
target_bot	Bot ID to attach the imported mission to. Defaults to the current bot from context.	String
target_bot_id	Deprecated alias for `target_bot`.	String
new_name	Custom name for the imported mission. Optional.	String
handle_conflict	How to handle name conflicts: `error`, `rename`, or `replace`. Default: `rename`.	String
repo_conflict	How to handle existing repos: `create_new`, `reuse_existing`, or `replace`. Default: `create_new`.	String

Example

import_mission(
    repo_id="uploads",
    file_path="exports/mission_123.zip",
    target_bot="sales_assistant",
    repo_conflict="reuse_existing"
)

`thread_cost`

Calculate token usage and cache-aware LLM costs for the current thread or any specified thread, with optional recursive child-thread aggregation. Parameters

Name	Definition	Format
thread_id	Thread ID to analyze. Defaults to the current thread from context.	String
target_bot	Optional bot ID for the thread. Defaults to the current bot. Only Eve can access other bots’ threads.	String
target_bot_id	Deprecated alias for `target_bot`.	String
include_children	Include child threads such as delegated or validation threads. Default: `true`.	Boolean
override_input_cost	Optional override for input-token pricing per 1M tokens. Applies to all models in the calculation.	Number
override_output_cost	Optional override for output-token pricing per 1M tokens. Applies to all models in the calculation.	Number
override_cache_read_cost	Optional override for cache-read pricing per 1M tokens. Applies to all models in the calculation.	Number
override_cache_write_cost	Optional override for cache-write pricing per 1M tokens. Applies to all models in the calculation.	Number

thread_cost returns separate input, output, cache_read, and cache_write totals. In summary.total_tokens, Genesis reports billable_input as prompt input minus both cache_read and cache_creation, floored at 0.

Example

thread_cost()

thread_cost(thread_id="chat_123", include_children=False)

thread_cost(
    thread_id="chat_123",
    override_input_cost=2.4,
    override_output_cost=12.0,
    override_cache_read_cost=0.24,
)

`compress_thread` (Internal / Support Only)

compress_thread is an internal function used by support workflows. Agents should use manage_thread_memory (part of context_tools, always available) for thread compression, status, and recovery.

See manage_thread_memory for the supported compression versions and fallback behavior. Thread memory management — compression, status checks, and recovery of compressed artifacts — is handled by manage_thread_memory in the context_tools group. Key actions:

Action	Purpose
`GET_STATUS`	Panel-style memory snapshot aligned with the context-state budget card
`COMPRESS`	Compress toward a target percentage (default 50%)
`SEARCH_MESSAGES`	Search visible, hidden, summary, raw, or artifact content within the current thread
`LIST_RECOVERABLE_BLOCKS`	Discover recovery handles for compressed blocks
`GET_COMPRESSED_TOOL_RESULT`	Retrieve compressed content or quote an uncompressed result still stored in the thread; see `manage_thread_memory` for paging details
`GET_COMPRESSED_PARAM_ARCHIVE`	Retrieve archived large tool parameters
`GET_HIDDEN_MESSAGES`	Retrieve logically hidden messages by durable `start_message_num` / `end_message_num` for V6, or by legacy handles when needed

Examples

manage_thread_memory(action="GET_STATUS")

manage_thread_memory(
    action="COMPRESS",
    target_percentage=50
)

manage_thread_memory(
    action="GET_COMPRESSED_TOOL_RESULT",
    result_tool_call_id="call_abc123"
)

manage_thread_memory(
    action="SEARCH_MESSAGES",
    query="French Russian",
    search_scope="hidden"
)

manage_thread_memory(
    action="GET_HIDDEN_MESSAGES",
  start_message_num=12,
  end_message_num=12
)

Compression artifacts include recoverable, recovery_kind, and recovery_hint metadata. For V6 history, use LIST_RECOVERABLE_BLOCKS and SEARCH_MESSAGES to find numeric start_message_num / end_message_num handles, then recover exact text with GET_HIDDEN_MESSAGES. Older summaries may still expose recovery_id as a legacy fallback. When GET_STATUS includes compression_history, each event can expose event_sequence (monotonic pass number) and compact raw diagnostics such as chat_model, thread_model_tier, thread_model_provider, summary_llm_model, summary_llm_engine, summary_rollup_fallback_used, and llm_summary_used_delta so you can see which models and summary paths were active for that pass.

Use Cases

Generate descriptive chat titles for better search and navigation.
Monitor platform health, component readiness, and external connectivity.
Schedule maintenance windows with user-facing messages and durations.
Adjust configuration keys safely and audit the changes.
Reclaim space by cleaning temporary files, old logs, or expired sessions.
Restart a misbehaving component after validating dependencies.
Maintain an administrative trail of system actions for compliance.
Recover a compressed tool result or a logically hidden message group without rerunning earlier work.

Workflow/How It Works

Assess current health with SYSTEM_STATUS and decide on next steps.
Communicate by enabling MAINTENANCE_MODE with a clear message and duration.
Change settings via SYSTEM_CONFIG to apply the desired configuration.
Clean using CLEANUP_TASKS with dry_run = true, then execute for real.
Restart selected components using SYSTEM_RESTART with explicit confirmation.
Verify recovery with SYSTEM_STATUS and disable maintenance mode.

Prefer dry_run for cleanups and staged config updates to reduce risk in production.

Integration Relevance

genesis_job_tools to coordinate or pause background jobs during maintenance.
system_stats_tools for performance verification before and after changes.
make_baby_bot to manage data agent availability during platform updates.
harvester_tools to align metadata crawls with maintenance windows.
All tools that rely on system configuration or uptime benefit from predictable scheduling.

Configuration Details

Use precise config_key names and document them in your runbook.
Maintenance messages should be concise and include a contact or status page.
Choose conservative retention_days to avoid premature log or session removal.
Component names for restarts should match deployed service identifiers.
Auto-title works best with a short, representative conversation_content when available.

Limitations or Notes

Administrative privileges are required for manage_system actions. export_thread instead follows the caller’s thread access permissions.
Maintenance mode affects all users, so schedule during low activity periods.
Some configuration changes may require a restart to take effect.
Cleanups are destructive unless you have separate backups.
Status checks can have overhead during peak load.
Cross-component dependencies may delay restarts or require ordering.

Output

Auto-Title: Proposed title, relevance score, and suggested alternatives.
Status: Overall health plus component details and connectivity results.
Maintenance: Mode status, banner message, and timing confirmation.
Configuration: Current values, change confirmations, and diffs where applicable.
Cleanup: Dry-run preview or executed results with counts and recovered space.
Restart: Acknowledgment, component recovery status, and timing.
Exports & Imports: export_thread, export_agent, and export_mission return the saved repo location plus download_url; export_thread also returns view_url and the selected export format. import_agent returns bot_id, bot_name, and warnings. import_mission returns project_id, project_name, thread_id, bot_id, repos_imported, query_results_imported, todos_imported, assets_imported, and warnings.
Thread Compression & Recovery: Agent-facing recovery uses manage_thread_memory, which returns panel-aligned status, before and after compression stats, searchable thread-history results, chunked value_chunk recovery with has_more and next_offset, and paginated hidden messages keyed primarily by durable message numbers/ranges for V6. The internal compress_thread helper backs those recovery flows for support workflows.
Errors: Actionable messages for permission issues, invalid keys, or failed operations.

​Overview

​Tool Name

​Purpose

​Key Features & Functions

​Input Parameters for Each Function

​AUTO_TITLE_CHAT

​SYSTEM_STATUS

​MAINTENANCE_MODE

​SYSTEM_CONFIG

​CLEANUP_TASKS

​SYSTEM_RESTART

​find_threads

​read_thread

​export_thread

​export_agent

​export_mission

​import_agent

​import_mission

​thread_cost

​compress_thread (Internal / Support Only)

​Use Cases

​Workflow/How It Works

​Integration Relevance

​Configuration Details

​Limitations or Notes

​Output

Overview

Tool Name

Purpose

Key Features & Functions

Input Parameters for Each Function

`AUTO_TITLE_CHAT`

`SYSTEM_STATUS`

`MAINTENANCE_MODE`

`SYSTEM_CONFIG`

`CLEANUP_TASKS`

`SYSTEM_RESTART`

`find_threads`

`read_thread`

`export_thread`

`export_agent`

`export_mission`

`import_agent`

`import_mission`

`thread_cost`

`compress_thread` (Internal / Support Only)

Use Cases

Workflow/How It Works

Integration Relevance

Configuration Details

Limitations or Notes

Output