DeCeroAI Solutions

Specialized AI services firm

Multi-Tool AI Agent Platform

Shipped an autonomous AI operator that lives in a company's Slack workspace, built on the open-source OpenClaw harness. 84 tools across 13 SaaS surfaces: it reads context, takes action, and reports back.

Headline number
84 tools, one operator
Role
Architected and built
Category
AI Agents & Runtimes
Updated

Architecture

  1. Slack workspacetag in any thread
  2. Agent runtime84 tools, 5 secrets
    Memory store
  3. Secret proxy sidecarlocalhost only, 12 secrets
  4. 13 SaaS surfacestrackers, Drive, email, code
    Key Vaultall 15 secrets by reference
Slack events reach the agent runtime, which calls SaaS APIs only through a localhost sidecar that holds the service secrets.

Key takeaways

  • 84 tools across 13 SaaS surfaces behind one conversational interface
  • Two-container design: the agent runtime never holds the service secrets
  • Per-user identity isolation, so the agent acts as a service identity and never as a person
  • Sandboxed coding sub-agent that writes code, runs tests, and opens pull requests

The situation

The firm wanted an agent that does work, inside the tools the team already lives in. Tag it in a Slack thread and it should read the project tracker, pull the meeting transcript, draft the document, send the email, and report back in the same thread. Most "AI assistants" stop at the chat window. This one had to operate.

What I built

I built the agent on the open-source OpenClaw harness in TypeScript and wired 84 tools across 13 surfaces: project trackers, source control, Slack, Google Drive, Docs, Sheets, Slides, email, calendar, meeting transcripts, and a self-hosted memory store for long-term context. A sandboxed coding sub-agent handles code changes and pull requests.

The deployment is the part I am proudest of. The main container runs the agent and holds only five secrets. A localhost-only sidecar proxies every service call and holds the other twelve, injecting auth headers so the service credentials never enter the agent's environment. All fifteen live in Key Vault and are injected by reference at deploy. The platform moved from Container Apps to Kubernetes as it grew, with the same isolation model.

The outcome

The agent runs real workflows end to end for a working team every day. The isolation design meant the security review focused on policy rather than on where the secrets were.

Stack

OpenClawTypeScriptNode.jsAzure Kubernetes ServiceAzure Key VaultPostgreSQLSlack Socket ModeAnthropic APIClaude Agent SDK

Related work

Same discipline, different system.

Security & Operations

Agent Security Architecture

Designed and documented the security model for an AI agent with private data access, external egress, and untrusted input, the worst-case surface for prompt injection. Controls that hold up.

Read the case study
Security & Operations

AI Agent Observability

Built operational observability for an LLM agent in production: heartbeat, users served, top tools, spend, errors, and deploys. Most teams ship the agent and then discover they have no idea what it is doing.

Read the case study
AI Agents & Runtimes

AI Content Engine with a Human Approval Gate

Turns a stream of public market chatter into original, compliance-checked stories with branded charts. Nothing reaches an audience until a human approves it in Microsoft Teams, and every approval is priced into the performance record automatically.

Read the case study

Need this built inside your organization?

One person, architecture to deployment. Tell me what is stuck and I will tell you where I would start.

Book a callAll work →