CHouseUI
Using CHouse UIAI agents

AI feature catalog

Every AI feature with its permission, tool contexts, template variables and built-in agent — generated from the server.

Source
Generated from packages/server/src/services/ai (feature contracts and seeds), so it always matches the code.

Every AI feature in CHouse UI, where it appears, what it needs, the template variables it hands its agent, and the agent it runs on after a fresh install. Rebind a feature or edit its agent in AI Governance › Assistant — see AI agents.

Needs is the permission a user must hold to use the feature; the route checks it before any agent runs. Tools may use is the run context the feature gives its agent's tools — an agent bound to a feature can only use tools whose context the feature provides:

ContextWhat the tools can do
ClickHouse sessionRuns on the signed-in user's ClickHouse connection, with their data access rules.
Fleet nodesRead-only system.* queries on the nodes the feature resolved.
Your CHouse accessCalls CHouse's own API as the chatting user — they see only what their permissions allow.

Features marked Evidence only gather their evidence in code and hand it over as ctx.evidence; their agents work from that and call no tools.

All features

FeatureWhereNeedsTools may useBuilt-in agent
ChatChatai:chatClickHouse session, Your CHouse accessCHouse Assistant
Suggest memoriesChatai:chatEvidence onlyMemory Suggester
Optimize querySQL editorai:optimizeClickHouse sessionSQL Optimizer
Debug querySQL editorai:optimizeClickHouse sessionSQL Debugger
Optimization pre-screenSQL editorai:optimizeClickHouse sessionQuery Evaluator
Fleet Doctor scanDoctordoctor:runFleet nodesFleet Doctor
Optimize a heavy queryDiagnosticsai:optimizeFleet nodesHeavy Query Optimizer
Diagnose a server errorDiagnosticsai:optimizeFleet nodesError Diagnostician
Diagnose part healthDiagnosticsai:optimizeFleet nodesParts Diagnostician
Diagnose a schema findingDiagnosticsai:optimizeFleet nodesSchema Diagnostician
Draft a scheduled queryDataOpsai:optimizeClickHouse sessionScheduled Query Drafter
Scheduled query preflightDataOpsai:optimizeClickHouse sessionScheduled Query Reviewer
Scheduled query briefDataOpsai:optimizeEvidence onlyScheduled Query Briefer
Diagnose a scheduled runDataOpsai:optimizeEvidence onlyScheduled Run Investigator
Assess a recovery planDataOpsai:optimizeEvidence onlyRecovery Planner
Recommend health checksDataOpsai:optimizeClickHouse sessionHealth Promise Advisor
Data health briefDataOpsai:optimizeEvidence onlyData Health Briefer
Diagnose a data health incidentDataOpsai:optimizeEvidence onlyHealth Incident Investigator
Tune health checksDataOpsai:optimizeEvidence onlyHealth Promise Tuner
Correlate incidentsDataOpsai:optimizeEvidence onlyIncident Correlator
Explain an incidentObservabilityai:optimizeEvidence onlyIncident Explainer
Compile a watcherObservabilityai:optimizeEvidence onlyWatcher Compiler
Draft table contextObservabilityai:optimizeEvidence onlyTable Context Drafter
Auto-triageObservabilityai_agents:manageFleet nodesIncident Triager
Read upgrade release notesObservabilityai:optimizeEvidence onlyUpgrade Notes Reader
Draft a dashboardDashboardsai:optimizeClickHouse sessionDashboard Drafter
Explain this spikeDashboardsai:optimizeClickHouse sessionSpike Explainer
Draft a skillAI Governanceai:optimizeEvidence onlySkill Author
Draft an agentAI Governanceai:optimizeEvidence onlyAgent Author
Draft a harnessAI Governanceai:optimizeEvidence onlyHarness Author
Draft a guardrailAI Governanceai:optimizeEvidence onlyGuardrail Drafter

Chat features

Chat

ID chat · needs ai:chat · free-form chat answer

The chat bubble. The bound agent answers when a thread has no agent picked; users can pick any top-level chat agent they may use.

Tools may use: ClickHouse session, Your CHouse access. Built-in agent: CHouse Assistant.

No template variables — chat agents render without any.

Suggest memories

ID suggest-memories · needs ai:chat · structured answer (JSON contract)

Chobi › chat: after a conversation, suggests up to three notes worth keeping for the person or the team; a person chooses what is saved.

Tools may use: Evidence only. Built-in agent: Memory Suggester.

VariableTypeValue
ctx.evidencejsonThe conversation, what memory already holds, and the rules (JSON).

SQL editor features

Optimize query

ID optimize-query · needs ai:optimize · structured answer (JSON contract)

SQL editor › Optimize: rewrites a SELECT for speed and memory, proven with a before→after EXPLAIN estimate.

Tools may use: ClickHouse session. Built-in agent: SQL Optimizer.

VariableTypeValue
ctx.querystringThe SELECT to optimize (trimmed).
ctx.additionalPromptstringExtra instructions from the user (trimmed; empty when none).

Debug query

ID debug-query · needs ai:optimize · structured answer (JSON contract)

SQL editor › Debug: explains why a query failed and returns a validated fix.

Tools may use: ClickHouse session. Built-in agent: SQL Debugger.

VariableTypeValue
ctx.querystringThe failed query (trimmed).
ctx.errorstringThe ClickHouse error message (trimmed).
ctx.additionalPromptstringExtra instructions from the user (trimmed; empty when none).

Optimization pre-screen

ID check-optimize · needs ai:optimize · structured answer (JSON contract)

SQL editor: silently decides whether a query is worth optimizing before offering Optimize.

Tools may use: ClickHouse session. Built-in agent: Query Evaluator.

VariableTypeValue
ctx.querystringThe query to evaluate (trimmed).

Doctor features

Fleet Doctor scan

ID fleet-scan · needs doctor:run · structured answer (JSON contract) · also runs in the background

Doctor › Scan (manual, scheduled or alert-triggered): reviews every node and deep-dives heavy queries into a health report.

Tools may use: Fleet nodes. Built-in agent: Fleet Doctor.

VariableTypeValue
ctx.hoursnumberInvestigation window in hours.
ctx.nodeCountnumberNumber of nodes scanned.
ctx.overviewjsonPer-node overview (pretty-printed JSON).
ctx.needsPlaybookbooleanTrue when the scan found a heavy or top-memory query worth optimizing.

Diagnostics features

Optimize a heavy query

ID optimize-log · needs ai:optimize · structured answer (JSON contract)

Query Logs › Optimize: rewrites one heavy query from the log, proven with a before→after EXPLAIN estimate.

Tools may use: Fleet nodes. Built-in agent: Heavy Query Optimizer.

VariableTypeValue
ctx.node.idstringConnection id of the node the query ran on.
ctx.node.namestringConnection name of that node.
ctx.peakMemorystringObserved peak memory, or "unknown".
ctx.querystringThe cleaned query text (first 8000 characters).

Diagnose a server error

ID diagnose-error · needs ai:optimize · structured answer (JSON contract)

Errors › Diagnose: explains one server error from system.errors and gives a concrete fix.

Tools may use: Fleet nodes. Built-in agent: Error Diagnostician.

VariableTypeValue
ctx.node.idstringConnection id of the node under investigation.
ctx.node.namestringConnection name of that node.
ctx.error.codestringError code, or "?".
ctx.error.namestringError name, e.g. TOO_MANY_PARTS.
ctx.error.messagestringLast error message, or "(none)".

Diagnose part health

ID diagnose-parts · needs ai:optimize · structured answer (JSON contract)

Parts › Diagnose: explains part and partition health of one MergeTree table.

Tools may use: Fleet nodes. Built-in agent: Parts Diagnostician.

VariableTypeValue
ctx.node.idstringConnection id of the node under investigation.
ctx.node.namestringConnection name of that node.
ctx.databasestringDatabase of the table.
ctx.tablestringTable name.

Diagnose a schema finding

ID diagnose-schema · needs ai:optimize · structured answer (JSON contract)

Schema Advisor › Diagnose: turns one column-level finding into a concrete ALTER TABLE fix.

Tools may use: Fleet nodes. Built-in agent: Schema Diagnostician.

VariableTypeValue
ctx.node.idstringConnection id of the node under investigation.
ctx.node.namestringConnection name of that node.
ctx.databasestringDatabase of the table.
ctx.tablestringTable name.
ctx.columnstringColumn name.
ctx.columnTypestringCurrent column type.
ctx.categorystringFinding category: nullable, oversized or compression.
ctx.sizeLinestringOne line with rows / on-disk / uncompressed bytes, or empty.

DataOps features

Draft a scheduled query

ID draft-scheduled-query · needs ai:optimize · structured answer (JSON contract)

Scheduled Queries › New with AI: turns intent into an editable draft.

Tools may use: ClickHouse session. Built-in agent: Scheduled Query Drafter.

VariableTypeValue
ctx.evidencejsonThe evidence object the feature assembled (JSON).

Scheduled query preflight

ID assess-scheduled-query · needs ai:optimize · structured answer (JSON contract)

Scheduled Queries › Editor › AI preflight: risks and improvements before saving.

Tools may use: ClickHouse session. Built-in agent: Scheduled Query Reviewer.

VariableTypeValue
ctx.evidencejsonThe evidence object the feature assembled (JSON).

Scheduled query brief

ID summarize-scheduled-query · needs ai:optimize · structured answer (JSON contract)

Scheduled Queries › AI brief: purpose, health, meaningful change and whether action is needed.

Tools may use: Evidence only. Built-in agent: Scheduled Query Briefer.

VariableTypeValue
ctx.evidencejsonThe evidence object the feature assembled (JSON).

Diagnose a scheduled run

ID diagnose-scheduled-run · needs ai:optimize · structured answer (JSON contract)

Scheduled Queries › Run › Diagnose: ranks causes of one failed or slow run.

Tools may use: Evidence only. Built-in agent: Scheduled Run Investigator.

VariableTypeValue
ctx.evidencejsonThe evidence object the feature assembled (JSON).

Assess a recovery plan

ID plan-scheduled-recovery · needs ai:optimize · structured answer (JSON contract)

Scheduled Queries › Recover: assesses a bounded backfill before it runs.

Tools may use: Evidence only. Built-in agent: Recovery Planner.

VariableTypeValue
ctx.evidencejsonThe evidence object the feature assembled (JSON).

Recommend health checks

ID recommend-health-promise · needs ai:optimize · structured answer (JSON contract)

Data Health › New with AI: recommends checks for one table from bounded aggregates.

Tools may use: ClickHouse session. Built-in agent: Health Promise Advisor.

VariableTypeValue
ctx.evidencejsonThe evidence object the feature assembled (JSON).

Data health brief

ID summarize-data-health · needs ai:optimize · structured answer (JSON contract)

Data Health › AI brief for one protected dataset.

Tools may use: Evidence only. Built-in agent: Data Health Briefer.

VariableTypeValue
ctx.evidencejsonThe evidence object the feature assembled (JSON).

Diagnose a data health incident

ID diagnose-health-incident · needs ai:optimize · structured answer (JSON contract)

Data Health › Incident › Diagnose: monitor failure or bad data, likely impact, safe next actions.

Tools may use: Evidence only. Built-in agent: Health Incident Investigator.

VariableTypeValue
ctx.evidencejsonThe evidence object the feature assembled (JSON).

Tune health checks

ID tune-health-promise · needs ai:optimize · structured answer (JSON contract)

Data Health › Tune: recommends threshold and cadence changes from history.

Tools may use: Evidence only. Built-in agent: Health Promise Tuner.

VariableTypeValue
ctx.evidencejsonThe evidence object the feature assembled (JSON).

Correlate incidents

ID correlate-health-incidents · needs ai:optimize · structured answer (JSON contract)

Data Health › Correlate: groups incidents that share a credible cause.

Tools may use: Evidence only. Built-in agent: Incident Correlator.

VariableTypeValue
ctx.evidencejsonThe evidence object the feature assembled (JSON).

Observability features

Explain an incident

ID explain-incident · needs ai:optimize · structured answer (JSON contract)

Observe › Incident › Explain: narrates the stored root-cause chain and drafts catalog fixes.

Tools may use: Evidence only. Built-in agent: Incident Explainer.

VariableTypeValue
ctx.evidencejsonThe evidence object the feature assembled (JSON).

Compile a watcher

ID compile-watcher · needs ai:optimize · structured answer (JSON contract)

Observe › Watch with AI: compiles a sentence into one Data Health check.

Tools may use: Evidence only. Built-in agent: Watcher Compiler.

VariableTypeValue
ctx.evidencejsonThe evidence object the feature assembled (JSON).

Draft table context

ID draft-table-context · needs ai:optimize · structured answer (JSON contract)

Context › Draft with AI: drafts a table's curated context from metadata and an aggregate-only profile.

Tools may use: Evidence only. Built-in agent: Table Context Drafter.

VariableTypeValue
ctx.evidencejsonThe evidence object the feature assembled (JSON).

Auto-triage

ID triage-incident · needs ai_agents:manage · structured answer (JSON contract) · also runs in the background

AI Governance › Auto-triage: investigates new incidents, alerts, failed scheduled runs and Doctor findings on its own, and proposes actions for approval.

Tools may use: Fleet nodes. Built-in agent: Incident Triager.

VariableTypeValue
ctx.triggerstringWhat happened, e.g. "Failed scheduled run".
ctx.titlestringThe subject's title.
ctx.evidencejsonWhat CHouse knows about the subject (pretty-printed JSON).
ctx.nodesjsonConnections the agent may query with query_node (id and name).

Read upgrade release notes

ID match-upgrade-notes · needs ai:optimize · structured answer (JSON contract)

Monitoring › Upgrades › Check: reads the breaking-change notes between two versions against the names the workload uses.

Tools may use: Evidence only. Built-in agent: Upgrade Notes Reader.

VariableTypeValue
ctx.evidencejsonThe evidence object the feature assembled (JSON).

Dashboards features

Draft a dashboard

ID draft-dashboard · needs ai:optimize · structured answer (JSON contract)

Dashboards › Ask AI: drafts a dashboard, or panels to add to one, from a goal; every panel is checked and dry-run before review.

Tools may use: ClickHouse session. Built-in agent: Dashboard Drafter.

VariableTypeValue
ctx.evidencejsonThe goal, the dashboard being added to, and the panel rules (JSON).

Explain this spike

ID explain-spike · needs ai:optimize · structured answer (JSON contract)

Dashboards › Explain this spike: reads a panel's selected range, the changes around it, slow queries, mutations and merges, and explains the likely cause.

Tools may use: ClickHouse session. Built-in agent: Spike Explainer.

VariableTypeValue
ctx.evidencejsonThe panel, the selected range and the evidence gathered around it (JSON).

AI Governance features

Draft a skill

ID draft-ai-skill · needs ai:optimize · structured answer (JSON contract)

AI Governance › Assistant › Skills › Draft with AI: drafts or refines a custom skill (SKILL.md and reference files) for review.

Tools may use: Evidence only. Built-in agent: Skill Author.

VariableTypeValue
ctx.evidencejsonThe request and the live registry facts the draft may use (JSON).

Draft an agent

ID draft-ai-agent · needs ai:optimize · structured answer (JSON contract)

AI Governance › Assistant › Agents › Draft with AI: drafts or refines a custom agent — prompt, tools, skills, subagents and settings — for review.

Tools may use: Evidence only. Built-in agent: Agent Author.

VariableTypeValue
ctx.evidencejsonThe request and the live registry facts the draft may use (JSON).

Draft a harness

ID draft-ai-harness · needs ai:optimize · structured answer (JSON contract)

AI Governance › Assistant › Harnesses › Draft with AI: drafts or refines a custom harness — a decision for every built-in tool — for review.

Tools may use: Evidence only. Built-in agent: Harness Author.

VariableTypeValue
ctx.evidencejsonThe request and the live registry facts the draft may use (JSON).

Draft a guardrail

ID draft-guardrail · needs ai:optimize · structured answer (JSON contract)

AI Governance › Guardrails › Draft with AI: turns a goal into a guardrail draft (kind, settings, stages, action) for review.

Tools may use: Evidence only. Built-in agent: Guardrail Drafter.

VariableTypeValue
ctx.evidencejsonThe goal, the guardrail kinds with their settings, and the drafting rules (JSON).

Built-in agents

The 35 agents CHouse installs. Steps are the agent's step budget; its recursion limit defaults to four times that, and at least 24. Without a timeout of its own, a run stops after 4 minutes for a feature and 2 minutes in the chat. See Editing agents.

SQL Optimizer

Rewrites a SELECT for speed and memory from its DDL and EXPLAIN plan (SQL editor › Optimize).

SettingValue
Slugsql-optimizer
KindFeature agent
Used byOptimize query
HarnessFocused
Tuning10 steps
Toolslist_databases, list_tables, get_table_schema, get_table_ddl, get_table_size, get_table_sample, run_select_query, explain_query, get_database_info, get_running_queries, get_server_info, search_columns, analyze_query, validate_sql, export_query_result, get_slow_queries
Skillsall built-in skills on demand

SQL Debugger

Explains why a query failed and returns a validated fix (SQL editor › Debug).

SettingValue
Slugsql-debugger
KindFeature agent
Used byDebug query
HarnessFocused
Tuning10 steps
Toolslist_databases, list_tables, get_table_schema, get_table_ddl, get_table_size, get_table_sample, run_select_query, explain_query, get_database_info, get_running_queries, get_server_info, search_columns, analyze_query, validate_sql, export_query_result, get_slow_queries
Skillsall built-in skills on demand

Query Evaluator

Quick, silent pre-screen: is this query worth optimizing? (SQL editor).

SettingValue
Slugquery-evaluator
KindFeature agent
Used byOptimization pre-screen
HarnessFocused
Tuning4 steps
Toolsanalyze_query, get_table_ddl, get_table_schema
Skillsall built-in skills on demand

Heavy Query Optimizer

Optimizes one heavy query picked from the Query Logs, proven with a before→after EXPLAIN estimate.

SettingValue
Sluglog-query-optimizer
KindFeature agent
Used byOptimize a heavy query
HarnessFocused
Tuning8 steps · 16,000 output tokens
Toolsquery_node
Skillsall built-in skills on demand; pinned clickhouse-playbook (reference.md)

Error Diagnostician

Explains one server error from system.errors and gives a concrete fix.

SettingValue
Slugerror-diagnostician
KindFeature agent
Used byDiagnose a server error
HarnessFocused
Tuning8 steps · 8,000 output tokens
Toolsquery_node
Skillsall built-in skills on demand; pinned clickhouse-playbook (reference.md)

Parts Diagnostician

Diagnoses part and partition health of one MergeTree table.

SettingValue
Slugparts-diagnostician
KindFeature agent
Used byDiagnose part health
HarnessFocused
Tuning8 steps · 8,000 output tokens
Toolsquery_node
Skillsall built-in skills on demand; pinned clickhouse-playbook (reference.md)

Schema Diagnostician

Turns one Schema Advisor finding into a concrete ALTER TABLE fix.

SettingValue
Slugschema-diagnostician
KindFeature agent
Used byDiagnose a schema finding
HarnessFocused
Tuning8 steps · 8,000 output tokens
Toolsquery_node
Skillsall built-in skills on demand; pinned clickhouse-playbook (reference.md)

Fleet Doctor

Reviews every node's health, deep-dives heavy queries and writes the Doctor report.

SettingValue
Slugfleet-doctor
KindFeature agent
Used byFleet Doctor scan
HarnessFocused
Tuning12 steps · 16,000 output tokens
Toolsquery_node
Skillsall built-in skills on demand

Incident Triager

Investigates new incidents, fleet alerts, failed scheduled runs and Doctor findings on its own, and proposes actions for approval.

SettingValue
Slugincident-triager
KindFeature agent
Used byAuto-triage
HarnessFocused
Tuning10 steps · 4,000 output tokens
Toolsquery_node, kill_query, propose_remediation
Skillsall built-in skills on demand

Dashboard Drafter

Turns a monitoring goal into a reviewed dashboard draft (Dashboards › Ask AI).

SettingValue
Slugdashboard-drafter
KindFeature agent
Used byDraft a dashboard
HarnessFocused
Tuning8 steps · 8,000 output tokens
Toolslist_databases, list_tables, get_table_schema, get_table_ddl, validate_sql
Skillsall built-in skills on demand

Spike Explainer

Explains what most likely caused a spike on a dashboard panel, from evidence gathered as the viewer (Dashboards › Explain this spike).

SettingValue
Slugspike-explainer
KindFeature agent
Used byExplain this spike
HarnessFocused
Tuning3 steps · 3,000 output tokens
Toolsnone
Skillsall built-in skills on demand

Guardrail Drafter

Turns a goal into a guardrail draft for review (AI Governance › Guardrails).

SettingValue
Slugguardrail-drafter
KindFeature agent
Used byDraft a guardrail
HarnessFocused
Tuning3 steps · 3,000 output tokens
Toolsnone
Skillsall built-in skills on demand

Memory Suggester

Suggests notes worth keeping after a Chobi conversation, for the person to save or dismiss.

SettingValue
Slugmemory-suggester
KindFeature agent
Used bySuggest memories
HarnessFocused
Tuning2 steps · 1,500 output tokens
Toolsnone
Skillsall built-in skills on demand

Scheduled Query Drafter

Turns an operator's intent into an editable scheduled query draft.

SettingValue
Slugscheduled-query-drafter
KindFeature agent
Used byDraft a scheduled query
HarnessFocused
Tuning7 steps · 4,500 output tokens
Toolslist_databases, list_tables, get_table_schema, get_table_ddl, analyze_query
Skillsall built-in skills on demand

Scheduled Query Reviewer

Preflight risk review of a scheduled query before it is saved.

SettingValue
Slugscheduled-query-reviewer
KindFeature agent
Used byScheduled query preflight
HarnessFocused
Tuning6 steps · 3,000 output tokens
Toolsanalyze_query, validate_sql, get_table_schema, get_table_ddl, explain_query
Skillsall built-in skills on demand

Scheduled Query Briefer

Operational brief for one scheduled query.

SettingValue
Slugscheduled-query-briefer
KindFeature agent
Used byScheduled query brief
HarnessFocused
Tuning3 steps · 2,500 output tokens
Toolsnone
Skillsall built-in skills on demand

Scheduled Run Investigator

Ranks the likely causes of one failed or slow scheduled run.

SettingValue
Slugscheduled-run-investigator
KindFeature agent
Used byDiagnose a scheduled run
HarnessFocused
Tuning5 steps · 4,000 output tokens
Toolsnone
Skillsall built-in skills on demand

Recovery Planner

Assesses a bounded backfill / recovery plan before it runs.

SettingValue
Slugrecovery-planner
KindFeature agent
Used byAssess a recovery plan
HarnessFocused
Tuning3 steps · 2,500 output tokens
Toolsnone
Skillsall built-in skills on demand

Health Promise Advisor

Recommends Data Health checks for one table from bounded aggregates.

SettingValue
Slughealth-promise-advisor
KindFeature agent
Used byRecommend health checks
HarnessFocused
Tuning8 steps · 5,000 output tokens
Toolsget_table_schema, get_table_ddl, run_bounded_aggregate
Skillsall built-in skills on demand

Data Health Briefer

Operational brief for one protected dataset.

SettingValue
Slugdata-health-briefer
KindFeature agent
Used byData health brief
HarnessFocused
Tuning3 steps · 2,500 output tokens
Toolsnone
Skillsall built-in skills on demand

Health Incident Investigator

Diagnoses a Data Health incident: monitor failure or bad data.

SettingValue
Slughealth-incident-investigator
KindFeature agent
Used byDiagnose a data health incident
HarnessFocused
Tuning5 steps · 4,000 output tokens
Toolsnone
Skillsall built-in skills on demand

Health Promise Tuner

Tunes noisy or insensitive Data Health checks from their history.

SettingValue
Slughealth-promise-tuner
KindFeature agent
Used byTune health checks
HarnessFocused
Tuning4 steps · 3,500 output tokens
Toolsnone
Skillsall built-in skills on demand

Incident Correlator

Groups Data Health incidents that share a credible cause.

SettingValue
Slugincident-correlator
KindFeature agent
Used byCorrelate incidents
HarnessFocused
Tuning4 steps · 3,000 output tokens
Toolsnone
Skillsall built-in skills on demand

Incident Explainer

Narrates an incident's stored root-cause chain and drafts catalog fixes.

SettingValue
Slugincident-explainer
KindFeature agent
Used byExplain an incident
HarnessFocused
Tuning3 steps · 3,000 output tokens
Toolsnone
Skillsall built-in skills on demand

Watcher Compiler

Compiles a sentence into one Data Health check.

SettingValue
Slugwatcher-compiler
KindFeature agent
Used byCompile a watcher
HarnessFocused
Tuning3 steps · 2,500 output tokens
Toolsnone
Skillsall built-in skills on demand

Table Context Drafter

Drafts a table's curated context from metadata and an aggregate-only profile.

SettingValue
Slugtable-context-drafter
KindFeature agent
Used byDraft table context
HarnessFocused
Tuning3 steps · 2,500 output tokens
Toolsnone
Skillsall built-in skills on demand

Upgrade Notes Reader

Reads ClickHouse's breaking-change notes for an upgrade against the names one workload uses (Monitoring › Upgrades).

SettingValue
Slugupgrade-notes-reader
KindFeature agent
Used byRead upgrade release notes
HarnessFocused
Tuning3 steps · 6,000 output tokens
Toolsnone
Skillsall built-in skills on demand

Skill Author

Drafts and refines custom skills (SKILL.md and reference files) from a goal or pasted runbook.

SettingValue
Slugskill-author
KindFeature agent
Used byDraft a skill
HarnessFocused
Tuning3 steps · 12,000 output tokens
Toolsnone
Skillsall built-in skills on demand

Agent Author

Drafts and refines custom agents — prompt, tools, skills, subagents, harness and tuning — from a goal.

SettingValue
Slugagent-author
KindFeature agent
Used byDraft an agent
HarnessFocused
Tuning3 steps · 12,000 output tokens
Toolsnone
Skillsall built-in skills on demand

Harness Author

Drafts and refines custom harnesses with a decision for every built-in tool.

SettingValue
Slugharness-author
KindFeature agent
Used byDraft a harness
HarnessFocused
Tuning3 steps · 6,000 output tokens
Toolsnone
Skillsall built-in skills on demand

ClickHouse Data

Read-only investigator for CHouse Assistant: databases, tables, columns, SQL, query performance, server state and charts.

SettingValue
Slugclickhouse-data
KindChat agent
Used bysubagent of CHouse Assistant
HarnessFocused
Tuning12 steps
Toolslist_databases, list_tables, get_table_schema, get_table_ddl, get_table_size, get_table_sample, run_select_query, explain_query, get_database_info, get_running_queries, get_server_info, search_columns, analyze_query, validate_sql, export_query_result, get_slow_queries, render_chart, generate_query, optimize_query
Skillsall built-in skills on demand

Access Auditor

Users, roles, permissions, data access policies, personal access tokens, SSO, and ClickHouse users and roles.

SettingValue
Slugaccess-auditor
KindChat agent
Used bysubagent of CHouse Assistant
HarnessFocused
Tuning8 steps
Toolswhoami, list_users, get_user, list_roles, get_role, list_permissions, list_data_access_policies, list_my_access_tokens, list_clickhouse_users, list_clickhouse_roles, get_sso_settings
Skillsnone

Operations Analyst

Scheduled jobs and runs, Data Health promises and incidents, alerting, observability incidents, remediation actions and investigation notebooks, Doctor reports, fleet health, query regressions, capacity, upgrade readiness, the observability collectors and custom dashboards.

SettingValue
Slugoperations-analyst
KindChat agent
Used bysubagent of CHouse Assistant
HarnessFocused
Tuning8 steps
Toolslist_scheduled_jobs, get_scheduled_job, list_scheduled_runs, list_health_checks, get_health_check, list_health_incidents, list_alerting, list_observe_incidents, list_remediation_actions, get_investigation_notebook, list_doctor_reports, fleet_snapshots, get_query_regressions, get_capacity, list_upgrade_assessments, get_upgrade_assessment, get_upgrade_status, get_collector_status, get_collector_schedule, list_dashboards, get_dashboard, get_dashboard_annotations
Skillsnone

Platform Auditor

Connections, AI models, CHouse's AI agents and their usage, cost, data sent to providers and guardrails, external agents and MCP settings, and the audit log.

SettingValue
Slugplatform-auditor
KindChat agent
Used bysubagent of CHouse Assistant
HarnessFocused
Tuning8 steps
Toolswhoami, list_connections, list_ai_models, list_ai_agents, list_ai_guardrails, list_ai_guardrail_triggers, get_ai_usage, list_ai_requests, get_ai_request, list_agent_sessions, get_agent_governance, audit_log
Skillsnone

CHouse Assistant

Chobi's assistant: answers, investigates with read-only investigators, and makes changes in CHouse and ClickHouse — each checked by its tier.

SettingValue
Slugchouse-assistant
KindChat agent
Used byChat
HarnessAssistant
Tuning16 steps · 240 s timeout
Toolslist_databases, list_tables, get_table_schema, get_table_ddl, get_table_size, get_table_sample, run_select_query, explain_query, get_database_info, get_running_queries, get_server_info, search_columns, analyze_query, validate_sql, export_query_result, get_slow_queries, render_chart, generate_query, optimize_query, whoami, load_toolkit, navigate, open_entity, fill_form, kill_query, create_saved_query, update_saved_query, delete_saved_query, share_saved_query, create_database, create_table, drop_table, drop_database, run_sql_change, create_scheduled_query, update_scheduled_job, pause_scheduled_job, resume_scheduled_job, run_scheduled_job, delete_scheduled_job, create_health_promise, watch_table, update_health_promise, pause_health_promise, resume_health_promise, run_health_check, acknowledge_health_incident, snooze_health_incident, delete_health_promise, propose_remediation, create_alert_rule, update_alert_rule, mute_alert_rule, test_alert_channel, post_alert_message, clear_alert_events, acknowledge_observe_incident, dismiss_suggestion, pin_table_criticality, add_notebook_cell, update_cost_rates, update_collector_schedule, start_codec_trial, run_doctor_scan, update_doctor_schedule, update_fleet_alert_config, run_upgrade_assessment, create_dashboard, update_dashboard, add_panel, update_panel, remove_panel, restore_dashboard_version, delete_dashboard, add_dashboard_note, delete_dashboard_note, share_dashboard, update_table_context, upsert_metric, delete_metric, create_user, update_user, deactivate_user, assign_role, create_role, update_role, delete_role, create_data_access_policy, update_data_access_policy, delete_data_access_policy, set_role_policies, create_clickhouse_role, update_clickhouse_role, draft_ai_agent, run_ai_eval, publish_ai_agent, update_my_preferences
Skillsall built-in skills on demand
SubagentsClickHouse Data, Operations Analyst, Access Auditor, Platform Auditor