Give your AI the knowledge to do more.

Start with internal reply drafts. A customer-facing chatbot also needs a chat interface, retrieval integration, access controls and escalation rules. Customer records must stay within the appropriate account boundary.

Check it works: Test ordinary questions, exceptions and questions with no answer in your sources. Check that private cases never appear in another customer’s response.

Connect or export the records your agent is allowed to use. Sending messages, updating a CRM and responding automatically require separate tools, permissions and review rules.

Check it works: Compare the draft with the source conversation. Verify prices, product claims and commitments before sending; do not infer agreement from silence.

Configure shared access, record ownership and an update process. Gigabrain does not automatically combine private workspaces or grant every teammate access to every source.

Check it works: Compare answers from two authorised users and test a user who should not see the account. Check the underlying storage permissions, not just instructions in a prompt.

You supply the catalogue and decision rules. Live stock, prices, recommendations on a website and order placement require connections to the systems that own those facts.

Check it works: Use cases with conflicting requirements and no suitable product. The agent should identify a poor fit, not force a recommendation.

Start with one project and a named owner for changes. When teammates share the knowledge, agree how edits are reviewed and which record is current.

Check it works: Include an old plan and a later amendment. Ask a new session to find both, identify the current decision and cite the change.

Provide transcripts or set up supported local speech transcription. Calendar access and automatic reminders depend on your connected tools. Review transcription errors in names and numbers.

Check it works: Check each commitment against its source. A discussion about an action is not necessarily an agreement to do it.

Set up shared storage access and an agent for each permitted user. Choose who maintains each subject. A shared folder is not a substitute for a permission model.

Check it works: Test with a new starter’s actual access. Check outdated policies, missing owners and confidential records before inviting the wider team.

Give your agent the appropriate repository and Gigabrain access. It must inspect current code and run the project’s tests; stored notes alone cannot establish runtime behaviour.

Check it works: Use a change where a plausible shortcut would violate a documented constraint. Verify that the agent retrieves the reason and that its implementation still passes review.

Prepare supported, readable source files and record dates and provenance. Database connections or exports need setup. Test retrieval on a representative collection before expanding it; capacity depends on the data and hardware.

Check it works: Use questions with known answers and relevant passages outside the obvious first document. Verify citations and coverage, not just the fluency of the summary.

Define the jurisdiction, source set and date range. Protect privileged and confidential material. This is research support, not legal advice or a guarantee that the library contains all applicable law.

Check it works: Have a qualified reviewer check the interpretation and currency. A verified quotation proves a match to the stored text, not that the text is authoritative or the conclusion correct.

Specify the audience, learning objectives and source permissions. Review subject accuracy and the suitability of examples; do not reproduce protected material without permission.

Check it works: Trace a lesson’s claims to the originals and ask a subject expert to review the sequence. A large collection is not necessarily a complete or balanced one.

Define the permitted purpose and access boundaries. Financial, legal and risk decisions need qualified review. Gigabrain is not a turnkey fraud detector, credit-decision system or compliance certification.

Check it works: Plant a known inconsistency in a test collection. Check that it is found and described accurately without treating a relationship as proof of wrongdoing.

Import permitted records and review the dependency model. Live inventory, forecasting, procurement execution and real-time monitoring require additional systems and integration.

Check it works: Compare the brief with a known operational change. Verify dependencies with the people responsible before acting on the result.

Choose examples you are allowed to reuse and approve saved preferences. Keep client-confidential sources separate. Gigabrain supplies context; you still review the writing.

Check it works: Check voice and structure, then verify every name, price and promise. Similarity to a good example does not make a new claim accurate.

Begin with a small set of non-sensitive records. Calendar changes, purchases and messages require connected tools and your chosen approval rules. Hosted agents may process material you share with them.

Check it works: Start a new session and ask it to recover a known decision. Check that it distinguishes recorded facts from suggestions and does not invent a deadline.

Configure each agent’s file or tool access. Save useful updates deliberately; Gigabrain does not automatically read every conversation. Team access and synchronisation need their own setup.

Check it works: Ask the same factual question in two configured agents and inspect their sources. The wording may differ; the underlying current record should be identifiable.