DECKLOG.
Knowledge operations for Microsoft 365 Copilot agents.
Prepare the knowledge layer before agents answer from SharePoint, OneDrive, policies, tickets, and contracts. DECKLOG checks permissions, source freshness, citations, and retrieval quality before rollout.
Make the corpus fit for agents.
Copilot answers are only as useful as the content and permissions behind them. DECKLOG turns scattered policies, runbooks, tickets, and contracts into a governed corpus with owners, freshness checks, and quality tests.
Live Topology
Active Policies
Event Firehose
LIVEDECKLOG fits around Copilot, Agent 365, and connectors.
Microsoft gives you the agent surface and governance primitives. DECKLOG prepares the knowledge sources those agents rely on, then measures whether the answers stay grounded after rollout.
The first pilot uses documented Microsoft paths: SharePoint and OneDrive, Copilot connectors, declarative agents, Teams, Purview, and Agent 365 where licensed.
Private answer surface for scoped pilots.
Some corpora should not be indexed into Microsoft 365 during the first pilot. DECKLOG can present the same answer shape in a tenant-owned private UI while the source strategy is decided.
A pilot wires that surface to your own scoped sources. Every answer names the document it came from, the section it was read from, and when that document was last changed, so an owner can check the answer against the source rather than trust the model.
There is no self-serve demo on this page. The surface is only meaningful against a real corpus with real permissions, so a walkthrough is a call: 30 minutes, your sources, your questions.
Book a DECKLOG walkthroughBuilt for Copilot agent rollouts.
Modules for the work around the agent: corpus cleanup, permission review, connector health, and answer-quality evidence.
Corpus Health Map
- Visualise knowledge sources, document owners, and freshness metrics
- Detect exact and semantic duplicates across SharePoint and Drive
- Flag conflicting policy versions before agents use them as context
Permission Exposure Scanner
- Simulate what specific users or groups can see after enabling Copilot
- Detect over-permissioned document libraries and open tenant links
- Validate ACL inheritance from source systems to the index
Citation Evidence Pack
- Production answers carry source citation requirements
- Capture document version, timestamp, hash, and ACL context
- Flag unsupported answers for SME review instead of inventing sources
Eval Workbench
- Golden set built with your subject matter experts
- Recall@K and nDCG checks for retrieval quality
- Groundedness, completeness, and refusal quality scoring
Connector Health
- Monitor sync latency across Microsoft Graph, Atlassian, and Google APIs
- Track webhook delivery failures and trigger automatic reconciliation
- Identify stale chunks or deleted documents still present in the index
Purview Bridge & Governance
- Map Microsoft Purview labels and DLP controls into the rollout plan
- Track source owners, review cadence, and remediation status
- Maintain an agent inventory with data scope and risk tier
Start with readiness. Prove one corpus. Operate the rollout.
DECKLOG starts narrow: one tenant, one business domain, one measurable answer set. Scope grows only after the first corpus passes readiness and evaluation.
Every path starts with scoping, because corpus size and risk decide the work.
Agent Readiness Audit
Best for
Tenants planning to enable Microsoft 365 Copilot or deploy custom agents, needing a factual risk assessment first.
- SharePoint and OneDrive permission exposure scan
- Oversharing and stale-site detection
- Purview label coverage analysis
- Corpus health report with remediation steps
- Identification of top 3 safe agent use cases
Outcome
Clear visibility into what Copilot will expose, what needs fixing, and which sources are ready for agents.
Scoped Pilot
Best for
Validating the value of a grounded, private agent on a specific knowledge corpus (e.g., HR policies or IT runbooks).
- Ingestion of 1-2 source systems (e.g., SharePoint + Confluence)
- Delivery of a Teams Agent or Copilot Connector
- Golden question set with 30 SME-validated test cases
- Citation evidence pack implementation
- Exit Kit: repo, manifests, runbook, eval dataset
Outcome
A working agent inside Teams answering questions with source citations, backed by quantitative quality metrics.
Operate Retainer
Best for
Production agent deployments requiring continuous evaluation, multi-system sync, and engineering SLAs.
- Continuous Connector Health monitoring and drift reconciliation
- Quarterly evaluation refresh with new golden questions
- Private RAG scoping for non-Microsoft 365 data
- Named AI engineer with SLA triage on quality regressions
- Monthly executive review of agent usage and ROI
Outcome
Reliable, cross-system knowledge operations. New-hire ramp shortens; senior staff stop answering repeat questions.
Looking for an implementation your team can own outright? Request DECKLOG Implementation scope
Microsoft-first. Custom only where needed.
Use Microsoft and prebuilt connectors first. Add custom Graph connectors only when the source is critical and no maintained connector fits.
Microsoft-first by default. Private RAG only when the source demands it.
The default pilot keeps Microsoft 365 as the system of record. Private retrieval, separate vector storage, and self-hosted models are later options for sensitive or non-Microsoft sources.
SharePoint, OneDrive, Teams, Copilot connectors, Purview, and Agent 365 stay as the primary platform. DECKLOG stores operational metadata, eval results, and evidence artefacts.
For sources outside Microsoft 365, the vector store and traces can run in the customer cloud account. This is scoped after the readiness audit, not assumed.
Self-hosted retrieval and generation remain an option for strict data-plane requirements. It is a later delivery path, not the default MVP.
Evidence packs mapped to DORA, NIS2, GDPR, and ISO 27001.
The readiness audit maps sources, owners, permissions, retention, and AI usage risks to the controls your team needs to review. Formal compliance statements stay inside the scoped deliverable.
Third-party ICT-risk evidence: model provider, hosting region, sub-processors, exit strategy, concentration analysis.
Technical baseline: PII redaction at boundary, MFA-bound query surface, audit logging, incident-response integration.
Right-to-erasure SLA documented; records of processing produced automatically with sub-processor + retention disclosure.
A.5 organisational + A.8 technological controls mapped per workflow with timestamped evidence chain.
Measured, reviewed, portable.
The pilot leaves behind artefacts your team can inspect: source inventory, golden questions, test results, agent manifests, and a runbook.
The golden questions, expected sources, refusal cases, and answer checks stay in the Exit Kit so your team can rerun them after corpus changes.
The pilot report shows source coverage, permission findings, stale content, connector health, and sampled answer quality. Runtime tracing is scoped per deployment model.
You receive the architecture notes, manifests, eval dataset, source map, and operating runbook. The next vendor can understand the deployment without starting from zero.
The questions AI and knowledge owners ask before purchase.
How does DECKLOG compare with Microsoft 365 Copilot?
DECKLOG prepares and operates the knowledge layer around Copilot. Microsoft provides Copilot, Agent 365, connectors, and governance primitives. DECKLOG turns those primitives into a readiness audit, source-owner cleanup, golden-question tests, and evidence packs for the agents you deploy.
Why do we need evals if we use Microsoft 365 Copilot agents?
Microsoft can only reason over the content and permissions it receives. If a stale 2023 policy is still discoverable, an agent can cite it. DECKLOG tests expected answers, expected sources, refusal cases, and permission boundaries against a golden-question set before rollout and after corpus changes.
Where does customer knowledge live?
The Microsoft-first pilot keeps Microsoft 365 as the system of record. DECKLOG stores operational metadata, eval results, connector health, and evidence artefacts. If a corpus needs private RAG outside Microsoft 365, storage location and model provider are scoped before implementation.
How does DECKLOG handle permissions on confidential documents?
The audit starts with source permissions, site ownership, sharing links, group access, and sensitivity labels. During a pilot, retrieval is scoped to approved sources. Permission-boundary questions become part of the golden-question set before a team gets access.
How long does a pilot take?
A typical Scoped Pilot takes 6 to 8 weeks. This includes the initial permission exposure scan, connecting 1-2 source systems, building the 30-question evaluation set with your experts, and deploying the agent to a test group in Teams. We deliver a complete Exit Kit at the end. You own the integration code, the eval dataset, and the deployment manifests.
What is the Exit Kit?
The Exit Kit gives your team the implementation map: agent manifests, source inventory, eval dataset, runbook, and operational decisions. The goal is simple: your tenant and your knowledge remain portable.
Make Copilot agents answer from the right sources. Prove it before rollout.
30-minute readiness call. We map your corpus, risk signals, and top three agent use cases, then recommend the smallest useful pilot.
Prefer written scope first? Email us