Positioning guide

AI workspaces are cockpits. ContextLattice is the memory infrastructure they were missing.

Odysseus, Open WebUI, LibreChat, Claude Desktop, Cursor, and custom MCP agents help people steer models. ContextLattice gives those surfaces a shared way to remember: one local-first write path, one retrieval contract, CLI-native handoff, Skills Index discovery, behavior provenance, and one memory spine that keeps getting smarter as the work grows.

Add memory to your agents Plug in an agent See the architecture

Plain claim

Not another cockpit. The black box, map, and nervous system beneath them.

The novelty is not a prettier chat window. It is shared agent infrastructure for the tools you already want to use: durable writes, sharp retrieval, bounded context, CLI workflows, Skills Index discovery, provenance trails, and deeper recall lanes when the work gets heavy.

Why this exists

Three problems appear once agents do real work.

Agents forget.

Chat history and workspace state are not durable memory. ContextLattice persists decisions, references, checkpoints, feedback, and task context so future agents can wake up with the thread already in their hands.

Long-context stuffing is expensive.

Dumping every file, note, transcript, and previous result into the prompt is panic with a bigger backpack. ContextLattice retrieves the smallest useful context package for the current task.

Every stack reinvents memory badly.

One app writes JSON, another keeps a SQLite cache, another ships a vector store with no write contract. ContextLattice gives local tools one memory and retrieval boundary.

Category clarity

The cockpit stays yours. The memory becomes shared.

Tool type	Examples	Primary role	Where ContextLattice fits
AI workspace	Odysseus, Open WebUI, LibreChat	User interface for model choice, chat, files, prompts, and local workflows.	Supplies persistent memory and scoped retrieval through HTTP or MCP.
Agent harness	Claude Desktop, Cursor, Codex, Claude Code, custom MCP agents	Actor that reads, writes, runs tools, edits repos, and executes tasks.	Provides checkpoints, recall, task handoff, and context-pack contracts.
Memory infrastructure	ContextLattice	Durable write path, retrieval policy, source fanout, topic rollups, Skills Index discovery, provenance trails, and agent-facing context contracts.	Acts as the shared memory spine beneath many cockpits and agents.

The contract

One local-first memory boundary for many agent surfaces.

Public local lite starts with topic rollups, Qdrant, CLI workflows, agent templates, and Skills Index discovery. Full/operator stacks can add pgvector, raw ledger, deeper async recall lanes, behavior provenance, and stronger reliability controls behind the same contract.

Write durability

Durable writes for decisions, checkpoints, notes, and evidence.
Fanout to configured stores without each agent inventing its own persistence path.
Compact payload contracts instead of full transcript dumping.

Retrieval quality

Topic rollups for fast local context.
Vector and structured stores for deeper recall.
Staged retrieval so slow sources can warm without blocking every answer.

Context reuse

Agent policy packs for session startup.
Context packs that stay bounded and task-scoped.
Handoff checkpoints so sister agents can continue without losing state.

Capability discovery

Skills Index search exposes quarantined capabilities on demand.
Agents do not need every low-probability skill loaded into the active prompt.
Tooling stays discoverable without turning startup context into a landfill.

Behavior provenance

Decisions, evidence, checkpoints, and agent/session context stay attached to the work.
Future agents can inspect why a state exists instead of trusting vibes from stale chat history.
Learning feedback improves recall while preserving the trail that made it useful.

Mental model

Workspaces and agents plug into the same memory spine.

Decision rule

Use ContextLattice when memory becomes infrastructure.

Use a workspace when...

You need a better chat surface.
You want to switch models visually.
You want document upload, prompt libraries, or local model controls.

Use an agent harness when...

You need code edits, shell execution, browser work, or tool calls.
You want an actor that can advance tasks instead of just chat.
You need local files and app-specific tools in the loop.

Use ContextLattice when...

You need state to survive across agents, tools, sessions, and days.
You need retrieval that improves without stuffing everything into prompts.
You need one write/search/checkpoint contract under many AI surfaces.