Keep your options.
Keep company knowledge useful when you change AI providers.Make company knowledge portable across supported models. Compare providers on the work that matters to you and bring the same background to each one.
GIGIAC BRAIN
You want useful AI without handing over every decision. Explore an approach built around reusable company knowledge, supporting sources and a choice of supported AI models. Start with one task and clear rules for what information can be used.
Your knowledge. Your model choices. Your rules.
All six documents.
Every request.
Another question. The same reading, all over again.
Custom payment terms require approval from the commercial lead before a proposal is shared with a customer.
Only this excerpt is added. The remaining full documents stay outside the request.
THREE QUESTIONS WORTH ASKING FIRST
Choose what the task needs to do, what information it may use and how you will check the result.
Make company knowledge portable across supported models. Compare providers on the work that matters to you and bring the same background to each one.
Follow important claims back to supporting sources. Give your team a clear way to check the reasoning before acting on an answer.
Define which information belongs in each workflow. Choose provider terms for training, retention and access that fit your business.
THE DECISIONS BEHIND AN AI DECISION
It is reasonable to ask these questions before bringing AI into company work. A useful starting point is one familiar task, a defined set of approved information and a way to check the result.
Start with a task you can explain and an outcome you can verify.
Each agent searches, reads and reconstructs the same knowledge.
GIVE AI USEFUL COMPANY CONTEXT
Think of a Context Capsule as a compact brief: useful facts, decisions and rules from approved company sources. Start with the essentials. Reach for supporting material when the task needs more.
Bring together the documents, conversations and code your agents are allowed to use.
Organize decisions, constraints and relationships. Keep each important claim linked to its source.
Create compact, reusable context with a path back to the full evidence.
Give authorized agents relevant context at the right depth, without repeating the whole briefing.
Review new decisions and useful results before adding them to shared memory.
pricing-policy.md / release-notes / project-decisions
MAKE THE ECONOMICS ADD UP
Start with the example, then enter your workload. Give existing caching its credit. Include deeper retrieval, maintenance and setup. The number that matters is the total cost of a successful task.
Expansion tokens use the uncached price. Current tokens should include all input already used by the baseline workflow.
Enter your prices and workload. Include retries in run counts.
YOUR COST ESTIMATE
43.2% lower total modeled cost, including setup allocation.
Every modeled cost.
One total to compare.
Cost per successful run
Current input cost = runs × current input tokens × blended input rate ÷ 1,000,000. Reusable input uses its own token count and cache coverage. Additional source cost = runs × expansion frequency × extra tokens × uncached input price ÷ 1,000,000. Expansion frequency is a fraction: 20% is 0.2. Extra tokens are added once, not included again in the base reusable context.
The blended rate weights cached and uncached prices by cache coverage. Output uses the same volume and price in both scenarios. Proposed total adds compile, refresh, retrieval, storage, platform, operating costs and setup divided by the selected months. Existing tool and labor costs are assumed unchanged unless added to operating costs.
Cost per successful run divides each total by runs × its success rate. Work per dollar compares those two unit costs. Break-even is the approximate monthly run count where proposed cost per successful run becomes no higher than the current workflow. It includes monthly setup allocation and assumes the same fixed overhead and per-run costs as volume changes. Different success rates can make break-even impossible even when total spending falls. Setup payback instead divides setup by monthly operating savings before setup allocation, without adjusting for quality. There is no payback when operating savings are zero or negative.
“I DON’T WANT TO BE LOCKED INTO A MODEL.”
Keep your company’s information separate from the model using it. Shared, portable context can reduce the background you prepare again when comparing supported providers. Choose the model that fits your workflow.
What we know. Why we chose it.
How the organization works.
The knowledge behind the answer.
Same memory. A different consumer.
Choose an example model role
Architecture decisions, project constraints and release policies.
Relevant capsules from the same shared memoryCONTEXT AT THE DEPTH THE TASK NEEDS
In this proposal example, start with the approved rate and rules. Open Working for more context, or Deep for the source excerpts. Then change the price and see the example update.
CONNECTED KNOWLEDGE
Prepare a proposal for Northstar Studio. Use the approved $2,000/month service rate. Keep the voice direct, helpful and specific. Custom terms require a human review. Sources: Pricing policy §2; Brand guide §1.
MEMORY WITH A PURPOSE
A debugging note and a security policy shouldn’t run on the same clock. Explore five memory tiers, from temporary working context to Core knowledge protected until an authorized human approves removal.
Keep working context available while a project is moving.
Rings show the time remaining in each tier. Core stays protected.
30 days until the inactivity threshold.
Retention is not freshness. A retained capsule can still need a source update or review.
Sources stay separate. Retiring derived memory does not delete original documents.
KEEP REUSABLE CONTEXT USEFUL
A new price or revised policy can make an old summary misleading. Keep reusable context connected to its sources: notice what changed, find the affected capsules and review conflicts before the next task.
An approved source is updated.
Find the capsules that depend on it.
Rebuild affected context and flag conflicts.
Serve the approved version to authorized agents.
“WHAT HAPPENS TO OUR DATA?”
Start with approved sources, clear access rules and a record of changes. Make data use part of the decision: who receives it, how long they keep it and what your provider agreement allows.
Trace important context back to supporting evidence.
START WITH ONE FAMILIAR WORKFLOW
Bring a recurring task and the questions holding you back. Let’s define the information, review process and measures for a focused first pilot.
Start with the work your agents repeat.
Compare cost, quality and completion time.
Define the workflow and the comparison.
Evaluate the result before expanding.
FOR YOUR ORGANIZATION
Tell us what you want AI to help with and what concerns you. If you are unsure about spend or providers, say so. We can explore whether an assessment makes sense.
FOR BUILDERS & TEAMS
Join the interest list for release updates and potential pilot opportunities for your team.
More valuable work.
For every dollar
you spend on AI.
INTELLIGENCE THAT COMPOUNDS
Start with a useful task, clear boundaries and results you can check.
BEFORE YOU CHOOSE
Reusable context targets the work agents repeat: finding background, reading the same sources and rebuilding a brief. Compare the full cost per successful task, including caching, context maintenance and implementation.
AI can still make mistakes. Source-linked context gives your team a way to check important claims against the underlying evidence. Keep human review where the decision calls for it.
Shared company context can carry over across supported providers, reducing the background you rebuild. Models differ in capability and output, so compare them on your actual tasks.
Your provider’s service, settings and agreement determine training use. Include training, retention and access terms when choosing the AI setup for your company.
The approach starts with your existing documents, code repositories and business systems. Your assessment identifies the sources and integrations your workflow needs.
The architecture separates source freshness from memory retention: refresh context when its evidence changes, and retire inactive capsules according to their tier. Original source documents stay separate.
Scope and pricing depend on your workload, integrations and security requirements. Start with an assessment to identify the repeated work and the value of a focused pilot.
Request an assessment with one recurring workflow and the context it needs. Or join early access for release updates and pilot opportunities.