ContextCapture
Ingestion surface
Ingests documents (PDF, DOCX, Markdown), videos, and audio. Source parsing, chunking, and metadata tagging feed LatticeCore data structures with structured, provenance-tracked content.
Explore ContextCapture → Platform Overview
One context supply chain. Built to run your organization's AI.
Seven products, three deterministic memory tiers, a six-layer agent framework, and end-to-end OTel observability — composed into a single governed context layer.
The Platform Thesis
Models provide intelligence. Context Lattice provides organizational understanding.
Generic AI models are general-purpose intelligence. They do not know which document is the source of truth, which exception matters, what your role-specific language means, or what agent actions are bounded inside your organization. That interpretation layer is what Context Lattice supplies.
The platform composes capture, curation, governance, packaging, activation, and feedback into a single supply chain. Every product surface contributes to and consumes from a shared context graph. Every completed workflow becomes a memory candidate that can promote into reusable organizational understanding.
Seven Products
Each product plays a distinct role. Together they form one context layer.
ContextBuilder
Contribution surface
Employee-submitted role questions, workflow friction points, internal language, edge cases, and demand signals. Community Q&A with maturity state progression from draft to verified.
Explore ContextBuilder →ContextCurator
Organization surface
Kanban-style context management links ContextCapture and ContextBuilder outputs into organized, governed context objects. Readiness scoring and agent work packet generation.
Explore ContextCurator →LatticeExplorer
Exploration surface
Explorer > Collections hierarchy for interactive browsing and querying of context atoms. Source-grounded answers, artifact generation, and context packs for export.
Explore LatticeExplorer →LatticeCore
Data layer
P1 CanonStore (exact-match quads for facts), P2 LatticeMem (session/role-aware memory), P3 Drop Corpus (semantic fallback). OTel-correlated trace IDs on every retrieval.
Explore LatticeCore →LatticeEngine
Execution layer
Context retrieval API, permission-aware context assembly, agent work packet delivery, P1 → P2 → P3 tier routing, memory writeback from verified agent outcomes, and evaluation harness.
Explore LatticeEngine →LatticeOperator
Orchestration & observation
Recommend and set up agentic orchestration architectures that utilize your context. Observe the functions and activities of all CL products from a unified observation plane.
Explore LatticeOperator → Three Memory Tiers
Deterministic memory first. Semantic search as a last resort.
Production knowledge systems require tiered memory. Context Lattice routes every context request through three tiers in precedence order — exact-match facts, role-aware aggregates, then semantic fallback.
TIER P1
CanonStore
Quad store for verified facts and relationships. SPARQL queries return exact ownership, dependencies, and source-of-truth precedence. Independent safety anchor for the Emotion Governor.
TIER P2
LatticeMem
Session and role-aware memory with permission-scoped retrieval. Holds curated atoms, employee contributions, decisions, and workflow scenarios. Graph-RAG over the company context graph.
TIER P3
Drop Corpus
Vector search over raw source material. Only consulted when deterministic tiers produce no match. Returns narrative context, never used as the source for ownership, status, or precedence answers.
Architectural Claim
Semantic similarity is not factual accuracy. A single vector store queried by embedding will return semantically proximate but factually incorrect answers for relationship and ownership queries. Production knowledge systems require deterministic tiers: exact-match quads for facts, aggregated statistics for trends, and semantic search as a last resort for narrative context.
Agent Framework Architecture
Six layers between human intent and agent action.
The agent framework is the independent variable. Benchmark rankings shift significantly from framework-only changes — making framework design a primary engineering lever, not a model-selection afterthought.
1. Safety
Unconditional constraints live in code. Action Governor, scope isolators, and the 4-tier intervention ladder enforce boundaries at the middleware — not the prompt — layer.
2. Orchestration
Ralph Loop continuation boundaries, TAO/ReAct cycles, and dual-channel cognition (Worker + Regulator) compose long-running agent sessions.
3. Execution
Tool invocation with progressive skill disclosure, context compaction, and counter-steering to prevent goal drift inside long sessions.
4. Context Assembly
Pulls from P1 / P2 / P3 memory tiers per request, with permission scoping and source-of-truth precedence applied before assembly.
5. Affective Sensing
Runs in parallel with memory retrieval, not downstream. Emotion vectors are re-sensed at every loop boundary rather than inherited, preventing state drift across long sessions.
6. Memory
P1 CanonStore + P2 LatticeMem + P3 Drop Corpus. Memory writeback promotes completed work to P2 candidates, then to P1 facts after human review.
Safety Principle
Unconditional constraints belong in framework code, not prompts. Safety guarantees that live in instructions can be reasoned around by the model. Context Lattice enforces action gates, scope isolators, and intervention ladders at the middleware layer — not the prompt layer.
Integration
Connects to the AI platforms you already use.
Context Lattice does not replace your AI assistants, coding agents, or copilots. It supplies governed context to them through context pack exports, retrieval APIs, MCP servers, and OTel-traced runtime feedback.
Context pack export (JSON, MDX, Markdown)
Retrieval API for runtime context
Native MCP server
Agent work packet generation
OTel trace correlation
Plane issue + commit trailer correlation