Repository instructions drift
Copied rules become stale, inconsistent, and difficult to update across repositories.
Geist Atlas · Deterministic context infrastructure
Geist Atlas gives every coding agent the same company context.
Turn the standards, policies, architecture decisions, runbooks, and business knowledge your team already maintains into controlled routing maps AI agents can retrieve on demand.
geist-atlas · active context system
The context gap
Agents can read code. They cannot infer how your business works.
AI coding agents make consequential engineering changes with a narrow view of your organization. They rarely know which policy is current, which architecture decision controls the task, or which operating rule applies. Atlas makes that knowledge an explicit, inspectable dependency.
Agents receive too much context
Large prompt dumps waste tokens and make the few rules that matter harder to find.
Retrieval changes invisibly
Opaque ranking can return plausible but different sources without a deliberate publication event.
Decisions lose provenance
Teams cannot reliably answer which policy or architecture version an agent used.
The Atlas system
Shared business context, controlled like infrastructure
Atlas preserves customer control from source selection to the exact context an agent receives.
FIG A.1
Approved sources
Connect the company knowledge your team already maintains. One source stays one traceable published document.
FIG A.2
Controlled routes
Group exact document versions into named routes your team defines, reviews, and publishes.
FIG A.3
Versioned delivery
Agents request only the route they need and receive the map and document versions with it.
How Geist Atlas works
Connect, publish, retrieve, revalidate
A deliberate path from company source material to current, traceable Atlas Context.
stage 01
Connect approved company knowledge
Keep source ownership where it already lives.
Atlas ingests the files and folders your team approves, normalizes supported content into deterministic Markdown, and preserves source paths, identifiers, and provenance. A source file remains a distinct document instead of disappearing into one opaque corpus.
- Google Drive folder ingestion for Google Docs and Markdown
- One source file maps to one traceable Atlas document
- Unsupported or failed files remain visible instead of silently vanishing
atlas sync · approved drive folder
→ scanning shared business context
policies/engineering-standards · Google Doc
architecture/service-boundaries · Markdown
operations/release-runbook · Google Doc
✓ 3 documents normalized
source URIs and Drive paths preserved
stage 02
Publish a controlled routing map
Treat company context like a versioned software dependency.
Atlas groups approved document versions into explicit routes such as engineering, security, architecture, or operations. Every document and map is content-addressed and immutable. The active pointer changes only after the complete new map has been written and verified.
- Named routes contain an ordered set of exact documents
- Identical content produces the same deterministic versions
- A failed publication leaves the previous active map usable
routing-map.json
{
"map_version": "527e4aed…",
"routes": {
"engineering": [standards, architecture],
"operations": [release-runbook]
}
}
verified → all documents written → active pointer switched
stage 03
Retrieve only the context the task requires
Give agents a route, not an indiscriminate knowledge dump.
The Geist Atlas skill teaches coding agents when company context matters. Agents list small route summaries first, select the relevant route, and retrieve complete documents within a visible byte budget. Required context failures stay visible.
- CLI output for agents, humans, and automation
- Explicit complete or incomplete retrieval status
- Map and document versions travel with the response
coding agent
$ geist-atlas context map
engineering · 2 docs · architecture and standards
operations · 1 doc · release procedure
$ geist-atlas context get engineering
complete: true · map: 527e4aed…
✓ 2 exact document versions loaded
stage 04
Revalidate current context without redownloading everything
Fast retrieval without quietly serving stale truth.
Atlas conditionally revalidates the active routing map, caches immutable documents by content version, and verifies cache integrity before use. Offline retrieval is explicit and reports when the cached map was published and last validated.
- Content-addressed local cache with integrity checks
- ETag revalidation for the active routing map
- Explicit offline mode with freshness and provenance
atlas cache status
active map 527e4aed…
revalidated with ETag · current
route manifest verified
document cache 2 / 2 hash-valid
offline results disclose publication and validation time
Atlas Context
The exact company context behind an agent decision
Atlas Context is not a loose collection of search results. It is a named route resolved against one published map and an ordered set of exact document versions.
That gives engineering teams a durable answer to five questions: which map was active, which route the agent requested, which documents it received, where those documents came from, and whether the response was complete.
Map
One immutable published snapshot
Route
A named, ordered document set
Document
Normalized content with source provenance
Version
A content hash that identifies exact bytes
Active pointer
The only mutable link in publication
Why teams need Atlas
Control and proof that copied instructions cannot provide
Geist Atlas replaces context drift with one governed delivery system for every supported repository and agent.
Capability
Geist Atlas
Copied instructions or opaque retrieval
Source control
✓Customer-approved files and folders
Whatever happens to be copied or ranked
Grouping
✓Explicit named routes with ordered documents
Duplicated instructions or opaque similarity
Updates
✓Immutable versions and an atomic active map
Silent drift across prompts and repositories
Retrieval
✓Smallest relevant route, bounded by design
Whole corpora or unpredictable snippets
Proof
✓Map version, document versions, and source provenance
Little evidence of what the agent used
Context domains
Route the knowledge that changes engineering decisions
Atlas is designed for governed company context that agents cannot safely infer from source code alone.
Engineering standards
Language rules, testing expectations, service conventions, and delivery requirements.
Architecture decisions
System boundaries, approved patterns, dependency constraints, and platform direction.
Security and policy
Data handling, access controls, compliance rules, and required review paths.
Business and product context
Customer promises, domain language, product constraints, and operating priorities.
Runbooks and operations
Release procedures, incident practices, ownership maps, and escalation paths.
Team onboarding
The shared knowledge engineers and agents need to work coherently from day one.
Atlas complements repository-native Agent Context Architecture. Repositories describe the local codebase; Geist Atlas supplies the governed organizational context shared across repositories and teams.
Geist Atlas FAQ
Questions about Atlas Context and controlled routing maps
The product model, boundaries, and operating behavior in plain language.
What is Geist Atlas?
Geist Atlas is deterministic context infrastructure for AI coding agents. It turns customer-approved company knowledge into explicit, versioned context routes that engineering agents can retrieve on demand.
What is Atlas Context?
Atlas Context is the current, traceable company knowledge delivered through a named route. A response identifies the routing-map version and the exact document versions used, so teams can see which context informed an agent.
How do controlled routing maps work?
A controlled routing map is a published snapshot that maps named routes such as engineering, security, or operations to an ordered set of approved document versions. New content creates a new version; publication changes the active map only after every referenced document is verified.
Is Geist Atlas a search engine or generic RAG system?
No. Atlas does not silently use semantic ranking to decide which company source is true. Teams explicitly control source inclusion and route membership, while agents retrieve the smallest relevant route for the task.
How does Geist Atlas integrate with coding agents?
The Geist Atlas CLI and agent skill let tools discover route summaries, retrieve a relevant route or exact document, detect incomplete or unavailable context, and preserve map and document versions when provenance matters.
What happens when a source document changes?
Atlas normalizes and hashes the changed content, creates a new immutable document version, builds a reviewable routing-map version, and atomically activates the map after validation. The previous active map remains usable if publication fails.
Give every agent the context your engineering team already knows.
Talk with Geist Labs about using Atlas for shared business context, explicit company routes, and traceable AI-assisted engineering work.