Clementino

A general-purpose 'AI colleague' spanning sales, finance, client ops, engineering, and recruitment, later split into a reusable toolkit plus Slack and desktop surfaces.

Company: Domu
Approach type: Agent
Work: Support, Finance ops, Coding, Recruitment, Customer success
Human involvement: Varies by scope: Work-product review, Continuous steering, Unknown. See each scope.
Invocation: Interactive, Scheduled
Interfaces: Slack, Desktop
Deployment stage: Deployed
Evidence strength: Detailed primary
Entry reviewed: 2026-09-16

Copy as MD Ask ChatGPT

How it works

Two named workflows: scheduled team jobs in Slack, and client-review preparation in the desktop toolkit.

Post the scheduled team jobs

Scheduled jobs such as the 5am sales standup, invoice approvals, QA sampling, and huddle ingestion post to Slack for team visibility^{[Source 1]}

Prepare a client review in the desktop toolkit

In Claude Cowork, the agent pulls call data from the warehouse and notes from the document store, then drafts the slides beside the user^{[Source 1]}

Where people stay involved

People review, approve, or guide the agent’s work here.

Reported: Customer-impacting actions run through team-visible Slack approvals; desktop work proceeds under continuous user steering.

Supervision level
Work-product review
A person reviews the draft or implementation.
Level 3 of 5
- employee request → approved customer-impacting action^{[Source 1]}
Supervision level
Continuous steering
A person pairs with the agent throughout execution.
Level 2 of 5
- individual works alongside the toolkit in Claude Cowork (client-review prep, spreadsheet cleanup, drafting long emails)^{[Source 1]}
Supervision level
Not classified
- scheduled jobs (5am standup, invoice approvals, QA sampling, huddle ingestion)^{[Source 1]}

How supervision levels work

Implementation details

Model

Claude (Anthropic SDK)^{[Source 1]}

Harness

Claude/Anthropic SDK wrapper; a reusable tool/skill/memory layer separated from the Slack and desktop interfaces^{[Source 1]}

Sandbox

Not reported

Tool access

~35 integrations organized into skills; specialist delegates per domain^{[Source 1]}

Knowledge

Four memory layers: conversation context, persistent facts, knowledge RAG, and live system state^{[Source 1]}

Context management

Four-layer memory separation; prompt size dropped substantially after splitting capabilities from the Slack layer^{[Source 1]}The prompt-size reduction that followed the split is a reported consequence, not a management mechanism.

Credentials

Customer-impacting actions gated behind team-visible Slack approvals^{[Source 1]}Approval gates govern customer-impacting actions; other credential handling is not documented.

Interfaces

slack
desktop

^{[Source 1]}

Mechanisms

Four memory layers

Transient conversation, persistent facts, knowledge RAG, and live system state; kept separate^{[Source 1]}

Reusable toolkit split

Capabilities separated from the interface layer so they power multiple surfaces^{[Source 1]}

Validation and failure handling

Not reported: No check of the agent's own work is documented beyond the approval gates cited with human involvement; QA sampling is a task the agent performs.

Reported observations

These results come from the sources. Notes below each result explain what it covers and any reported limits.

Headline claim

~35 integrations organized into skills^{[Source 1]}

Reported by: Domu
Scope: Clementino toolkit integration modules organized into skills

The source does not report the denominator of this figure.

About this result

Category: Implementation scale
Evidence type: Reported measurement
What it covers: Clementino toolkit integration modules organized into skills

Lessons

Domu separated Slack handling from Clementino's tools, skills, memory, and core, then reused the toolkit in Claude Cowork.^{[Source 1]}

Clementino shares conversation context, persistent facts, retrieval knowledge, and live system state across its Slack and desktop interfaces.^{[Source 1]}

Clementino routes customer-facing actions through visible Slack approvals, including requests started from the desktop.^{[Source 1]}

Back to agents

Improve this record

Give your coding agent the context and instructions to prepare a source-backed update.

Suggest on GitHub

Paste the prompt into your coding agent, then describe the correction or add a public source.

Preview prompt

Help me update the Internal Agents Map record for Clementino at Domu.

Repository: https://github.com/steel-experiments/internal-agents-map
Record: data/agents/domu-clementino.yaml
Published record: https://internal-agents.com/agents/domu-clementino
Readable context: https://internal-agents.com/agents/domu-clementino.md

Use the correction or public source I provide in this conversation. If I have not supplied either, ask what I want changed before editing.

Use an existing checkout of this repository, or clone it if needed. Read the current record from the repository. Follow AGENTS.md, CONTRIBUTING.md, and the relevant instructions in data/schema.md.

Verify that the evidence supports the requested change. Preserve uncertainty and distinguish reported facts from catalog judgments. Follow the repository process for capturing new sources. If evidence is insufficient, explain the gap instead of inventing details.

Make a focused change on a branch. Regenerate derived files and run the required checks. Do not edit generated content by hand.

Show me the change, supporting evidence, and check results, including any checks you could not run. Ask whether to submit a pull request unless I already requested one. Do not merge the pull request. If you cannot edit the repository, prepare a source-backed issue draft.

Sources and research details

Citations link to the original publisher. Each source also keeps a preserved copy in the repository, so a changed or removed page stays checkable.

Why we split our internal agent in twohttps://domu.ai/blog/why-we-split-our-internal-agent-in-twoEngineering blog · First party · Last source verification: 2026-08-31
Preserved copy

Question coverage and scope

purpose: Reported
workflow: Reported: The post names exactly two workflows; the memory layers and the toolkit split are reusable mechanisms, not runs.
human involvement: Reported: Customer-impacting actions run through team-visible Slack approvals; desktop work proceeds under continuous user steering.
implementation: Reported
validation: Not reported: No check of the agent's own work is documented beyond the approval gates cited with human involvement; QA sampling is a task the agent performs.
observations: Reported
lessons: Reported

Research details for every claim on this page

Summary
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsWhy we split our internal agent in two
Headline claim
Statement type
Metric
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
Reported by
Domu
Scope
Clementino toolkit integration modules organized into skills
Denominator
Not reported
Method
Not reported
Observation date
Not reported
- SupportsWhy we split our internal agent in twoPreserved content.md, lines 58–60
Harness
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsWhy we split our internal agent in two
Model
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsWhy we split our internal agent in two
Interfaces
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsWhy we split our internal agent in two
Tool access
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsWhy we split our internal agent in two
Knowledge
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsWhy we split our internal agent in two
Credentials
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsWhy we split our internal agent in two
Context management
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsWhy we split our internal agent in two
Four memory layers
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsWhy we split our internal agent in two
Reusable toolkit split
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsWhy we split our internal agent in two
Post the scheduled team jobs
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsWhy we split our internal agent in twoPreserved content.md, lines 70, 78
Prepare a client review in the desktop toolkit
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsWhy we split our internal agent in twoPreserved content.md, line 80
Key observation
~35 integrations organized into skills
Duplicate representation of Headline claim. Identical statement, scope, and locator as the headline: the same about-35 integration count.
Statement type
Metric
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
Reported by
Domu
Scope
Clementino toolkit integration modules organized into skills
Denominator
Not reported
Method
Not reported
Observation date
Not reported
- SupportsWhy we split our internal agent in twoPreserved content.md, lines 58–60
Lesson
Statement type
Fact
Provenance
Reported
Confidence
Medium
Confidence reason
The refactor and both interfaces are described directly. A reported prompt reduction accompanied the change, but its cause is left for a later post.
- SupportsWhy we split our internal agent in twoPreserved content.md, lines 54-72,98
Lesson
Statement type
Fact
Provenance
Reported
Confidence
Medium
Confidence reason
The toolkit section enumerates four memory layers stored in common tables; the two surfaces use that shared memory.
- SupportsWhy we split our internal agent in twoPreserved content.md, lines 64-72,90
Lesson
Statement type
Fact
Provenance
Reported
Confidence
Medium
Confidence reason
The author explicitly names invoices, outbound emails, and contract changes and says the initiating surface does not bypass the approval path.
- SupportsWhy we split our internal agent in twoPreserved content.md, lines 78,92
Operating model assessment
Statement type
Inference
Provenance
Catalog judgment
Confidence
High
Confidence reason
The source documents explicit human approval for customer-impacting actions.
Observation date
2026
- SupportsWhy we split our internal agent in two
Operating model assessment
Statement type
Inference
Provenance
Catalog judgment
Confidence
High
Confidence reason
The source describes the author working alongside the agent in Cowork and watching what happens in one place, with no approval gate described.
Observation date
2026
- SupportsWhy we split our internal agent in twoPreserved content.md, lines 76, 80 (two workflows; work watched in one place as it happens)
Operating model assessment
Statement type
Inference
Provenance
Catalog judgment
Confidence
Unverified
Confidence reason
The source lists scheduled jobs that post without a person starting them, but does not describe whether anyone reviews their output.
Observation date
2026
- SupportsWhy we split our internal agent in twoPreserved content.md, lines 70, 78 (scheduled jobs post at 5am; review, if any, not described)