Context Atlas
The domain-neutral foundation for anyone who needs controlled context for AI work.
- Local viewer and pack builder
- Shared data contract and templates
- Review, source and correction rules
← All products
Product Five | AI context toolkit
Context Atlas
A downloadable toolkit that helps you organise the facts, decisions and source files an AI assistant should use for one specific task.
Review information before it reaches the AI, connect related facts, then export a focused Markdown or JSON context pack instead of sharing your entire workspace.
Build a focused AI context pack with its sources, review status and correction history attached.
What you receive
- Working local viewer, schema and validation CLI
- Nine templates, three workspaces and finished result files
- 18-page guide, interoperability recipes and release proof
Runs from the downloaded folder. No hosted account is required.
£39 launch priceOne-time purchase
- 42 buyer files
- 18 release tests
- 3 worked examples
- 18-page guide
Interactive example
See one trusted fact become an AI context pack.
All recordsAll 7 graph items are visible.Nothing is hidden. This includes the provisional AI output and the saved correction.
EXAMPLE WORKSPACE / NORTHSTAR LAUNCH7 graph items
Use Product 5 and Product 6 together
Review the project first. Then build one bounded developer task.
In this example, six reviewed Product 5 sources become one Product 6 team-invitation pack. Test and migration evidence returns to Product 5 for review.
- 1Product 5Accept the feature brief and role policy.
- 2Product 6Trace the implementation and unresolved migration.
- 3Developer or coding agentImplement inside the pack's paths and guardrails.
- 4Back to Product 5Review tests and migration evidence before truth changes.
A practical example
Prepare a launch brief without re-explaining the project.
A fictional product team needs an AI assistant to prepare a release-readiness brief. Context Atlas turns reviewed meeting notes, decisions and constraints into one focused context pack for that task.
- Step 1Add the source files
The meeting note records that the beta date moved after an accessibility review.
Source attached - Step 2Review the important facts
A human confirms the new date, source quality, confidence and sensitivity before use.
Human approved - Step 3Build the context pack
Four relevant records are included. Two unrelated records stay outside this task.
Scope recorded - Step 4Check the AI output
The result is reviewed, corrected if needed and saved with a clear audit receipt.
Correction history kept
The Context Atlas product family
Product Five supplies the shared toolkit. Products Six and Seven add a job-specific layer.
All three use the same review rules, source tracking, typed relationships and portable context-pack format.
Context Atlas for Developers
Adds repositories, requirements, decisions, issues, tests, builds and releases for coding-agent workflows.
Buyer outcome: an AI project brain for software work.Open the developer product pageContext Atlas for Founders
Adds company, offers, customers, policies, goals, experiments and campaigns for business-AI workflows.
Buyer outcome: an AI company brain for running the business.Open the founder product pageProducts Six and Seven are separate editions with their own onboarding, examples and buyer outcomes. They do not replace the Product Five foundation.
See what the toolkit includes
Build a focused AI context pack with its sources, review status and correction history attached.
1 of 5
Product FiveShared foundation
Product manifest
- Contract v0.1
- 42 buyer files
- 18 tests passed
- 18-page guide
- Local-only interactive proof
Interactive product proof
Give the AI only the reviewed context needed for this task.
Walk a fictional source through the shared Product Five lifecycle. This stays in your browser and does not call an AI provider.
Local walkthroughNo provider call | No automatic trust
WORKSPACENorthstar Launch
PROFILENeutral core
CAPTURE 014
Beta date moved after accessibility review
The public beta target moved from 28 August to 30 August so the keyboard-navigation fixes can be verified before release.
- Source
- meeting-notes/2026-08-26-release-review.md
- Owner
- Mara Chen | fictional fixture
- Suggested links
- Public beta | Accessibility gate | Release checklist
HUMAN REVIEW
Should this source enter trusted context?
The note remains attached to every object and relationship created from it.
The date is a target, not a confirmed release commitment.
Safe for this fictional workspace; not promoted outside it.
TASK SCOPE
Choose exactly what the workflow may know.
3 included | 1 excluded with a reason
NEUTRAL EXPORT
Task pack ready for a declared capability.
{
"contract_version": "0.1",
"task": "Prepare the release-readiness pass",
"trust_policy": "reviewed-only",
"included": ["beta-target", "accessibility-gate", "release-checklist"],
"excluded": [{ "id": "customer-research", "reason": "not required for this task" }],
"adapter": { "target": "neutral-files", "status": "local export only" }
}Adapter boundary
Markdown + JSON are the current neutral path. ChaseOS and other runtimes remain untested adapter declarations.
OUTPUT VERIFICATION
The agent draft does not become memory automatically.
“Northstar Launch will release publicly on 30 August after the accessibility work is complete.”
Why it needs correction
The source says 30 August is a target, not a confirmed public release date. “Will release” overstates the reviewed evidence.
HUMAN CORRECTION
Save better memory without erasing its lineage.
PROOF RECEIPT
The context changed. The history stayed visible.
- 01Source captured
No project truth changed.
- 02Accepted with provenance
One source and three typed links retained.
- 03Task pack scoped
3 objects included; exclusions recorded.
- 04Output challenged
The draft overstated a target as a commitment.
- 05Human correction saved
Target remains conditional on keyboard-navigation verification.
What is included
Everything you need to build, review and export trusted AI context locally.
- A versioned shared-object contract with stable IDs, types, review status, provenance, ownership, dates, confidence, sensitivity and notes
- Typed relationship records and reviewed-memory states that stay inspectable instead of disappearing inside a prompt
- Task-sized Markdown and JSON context-pack formats with explicit inclusions, exclusions, sources and limitations
- Profile-pack and adapter declarations so vertical editions can add their own objects, onboarding, examples and exporters
- A human-owned lifecycle for capture, review, connection, export, use, verification, correction, expiry, archive and deletion
Purchaser delivery contract
Everything in the download, explained before you buy.
Version 1.0 is a working local toolkit. These are the exact deliverable groups in the customer ZIP and the job each one performs.
Local viewer + pack builder
viewer/index.htmlInspect the graph, review records, scope one task and export JSON or Markdown without a hosted account.
Portable core contract
contract/context-atlas-core.schema.jsonKeep sources, objects, typed relationships, review history and context packs in inspectable JSON.
Zero-dependency CLI
tools/context-atlas.mjsCreate, inspect, validate and export workspaces repeatably with Node.js 20 or newer.
Nine practical templates
templates/Start from explicit workspace, record, relationship, review, pack, profile and adapter structures.
Three completed workspaces
examples/Learn from the neutral walkthrough and see how the Developer and Founder editions reuse the same core.
Finished result outputs
results/neutral-launch/Open the exact Markdown and JSON pack produced by the example before creating anything yourself.
Two profile declarations
profiles/Extend the vocabulary for Products 6 and 7 without weakening the shared trust contract.
Adapters + other-system guide
adapters/ + docs/USING-WITH-OTHER-SYSTEMS.mdUse tested neutral files with assistants, agents, graphs and automation while keeping untested paths explicit.
18-page illustrated guide
manual/context-atlas-guide.pdfFollow the trust loop, viewer, CLI, privacy rules, family boundaries and troubleshooting visually.
Manifest, proof + hashes
manifest.json + PROOF-REPORT.mdVerify every buyer file, package claim, limitation and the exact Version 1.0 archive identity.
Actual result included
The finished result is a focused task pack an AI can actually use.
The included Northstar example turns a reviewed workspace into one release-readiness pack. The same result is delivered as readable Markdown and structured JSON.
PACK: RELEASE READINESSReviewed
Prepare a release-readiness review without publishing or treating the target date as confirmed.
- Included
- 4 reviewed records
- Connected
- 3 typed relationships
- Provenance
- 2 cited sources
Constraints carried into the task
- Do not publish or deploy.
- Treat 30 August as a target, not a confirmed release date.
- Keyboard navigation and focus order must pass before beta.
UncertaintyNo production deployment or buyer acceptance has been tested.
ExcludedCustomer research is left out because it is not needed for this task.
Use it with other systems
Portable files first. Tested integrations second.
Context Atlas does not require one AI provider or graph platform. Export a bounded Markdown or JSON pack, then use only the capabilities the receiving system has actually demonstrated.
AI and chat assistants
Attach the Markdown pack to one task. The assistant receives reviewed facts, sources, constraints, uncertainty and exclusions instead of a full data dump.
Coding agents
Place a task-sized pack in the agent's approved work folder. Product Six adds developer records, onboarding, fixtures and tested coding-agent recipes.
Business AI workflows
Export only the company context needed for the task. Product Seven adds founder records, onboarding, fixtures and business-specific recipes.
Knowledge graphs
Map nodes and typed edges while preserving stable IDs, sources, trust state, sensitivity, confidence and dates. Declare any fields the target drops.
Automation and version control
Validate before use, fail closed on broken references, and retain the workspace and task pack as ordinary files when the sensitivity allows it.
ChaseOS
Use the included architecture-fit declaration as an education path only. A tested ChaseOS or Studio adapter is not included in Version 1.0.
Products Six and Seven extend this same foundation. They add distinct onboarding, records, examples, screenshots, recipes, exporters and buyer outcomes. They are not folders renamed from Product Five.
Current build
Version 1.0.0 | Contract v0.1- 42 buyer files
- 18 release tests
- 3 worked examples
- 18-page guide
How the system works
The same four-step process powers every Context Atlas edition.
The Developer and Founder editions change the records, examples and buyer outcome. The review process and portable export format stay consistent.
- 01
Add and review
Bring in a source, then decide whether its facts are trusted, need correction or should stay out of the workspace.
- 02
Connect related facts
Record why a source supports a decision, constraint, goal or other piece of useful context.
- 03
Build the context pack
Export only the reviewed facts needed for one task, with sources, exclusions and limits included.
- 04
Check and correct
Review the AI output, save any correction with a receipt, then expire or archive stale information.
Interoperability contract
Compatibility is tested and stated clearly.
Neutral files work first. The buyer guide shows practical use with assistants, coding agents, business AI, graphs and automation. A named integration becomes compatible only after its exact target and capabilities are tested.
Markdown + JSON files
Release-tested in Version 1.0The CLI and local viewer create inspectable task packs without a hosted service or ChaseOS dependency.
Contract v0.1 examples
Three workspaces validatedNeutral, Developer and Founder examples pass the shared checks with zero dangling edge, source, review or context-pack references.
ChaseOS / Studio path
Optional exemplar; adapter untestedThe governed-system path is educational until a versioned adapter and direct acceptance evidence exist.
Other graphs and agent runtimes
Capability declaration requiredNo universal compatibility claim is made. Each target needs its own tested adapter declaration.
Build evidence
The local toolkit and its core contract have been tested.
The local product and buyer archive demonstrate capture, human review, graph connection, focused export, correction and audit history with fictional public-safe data. The clean-extracted package passes its declared file, validation and adapter paths. Named external runtime execution, paid-buyer delivery and customer outcomes remain separate proof states.
Try the guided context-pack example →- 42
- files in the buyer package
- 18/18
- release tests passed
- 3
- valid example workspaces
- 18
- rendered guide pages
Before you buy
What Product Five does and what it does not do.
Is Context Atlas a generic second brain?
No. The shared core is built around reviewed objects, typed relationships, provenance, task-sized exports and correction receipts rather than collecting everything into an undifferentiated archive.
Does it require ChaseOS?
No. Product Five is standalone and local-first. ChaseOS can become an optional governed-system adapter after direct, versioned testing; it is not a hidden dependency.
Are Developer and Founder objects built into the core?
No. They are profile layers. The core keeps object types open so later editions can own their own fixtures, onboarding, exporters and buyer outcomes.
Does the current preview run an AI agent?
No. It demonstrates the context governance path using fictional local data. Real runtime execution and output verification require separate adapter evidence.
Can I use it with an AI tool I already have?
Yes through the tested neutral file path when the tool accepts Markdown or JSON. Start with one task-sized pack and verify the output. Named tool compatibility, automatic import and writeback require their own tested adapter evidence.
Need this operating knowledge applied to your own system?
See ChaseInTech services →