Drop documents into a folder and search them by meaning from chat, prompts, workflows, or any external MCP client. Aisle handles extraction, indexing, and embeddings in the background. You skip the vector database.
Aisle splits every document into atomic claims, one fact per statement. That structure lets the model pull the top 20 facts ranked by similarity for a given query and trace each one back to its source document, which prevents hallucinated answers. Vector search runs across those atomic claims, so a query for “unhappy customers” still returns memories about cancellations.
Read about propositionsOnce a memory exists in a folder, the same folder feeds chats, projects, prompts, and workflows. No separate sync step. No per-surface configuration.
See how memories are usedYou can have one workflow create a memory and another workflow update it the next time it runs. A customer profile gets new notes appended every time a support ticket comes in. A research document grows as new findings are added. The same document stays available across runs without an external database.
See workflow operationsYou can attach key/value metadata to every memory. A workflow looking for active billing tickets queries status: active and category: billing and gets back exactly those memories. Comparison operators and existence checks are also supported, so the query layer is precise enough to drive workflow logic.
You can expose any memory folder as an MCP server. Once enabled, an external agent can call seven tools (vector_search, create_memory, update_memory, query_memories, find_memory, list_records, and get_propositions), the same operations a workflow uses internally. Each folder gets its own bearer token, so access is scoped to the folder you choose.
A memory folder works as a knowledge base for any team that builds with AI. The examples below cover both customer-facing roles and developer workflows.
Use the previous and next buttons or swipe horizontally to move between use cases.
Customer profiles, contracts, call notes, and product feedback live in one folder. Sales and CS teams chat with the folder to find the right context for any conversation, without paging anyone.
Your employee handbook, HR policies, benefits docs, and onboarding guides go into a folder. New hires chat with it to answer questions like "how do I file expenses?" without messaging an HR partner.
User research reports, interview transcripts, and competitive analysis sit in a memory folder. Anyone on the team can chat to find a quote, trace a finding to its source, or surface trends across studies.
Product docs, troubleshooting guides, and known-issue notes live in one folder. Support agents chat with the folder during a ticket and get the right answer along with the source citation.
Memory works alongside the prompts, workflows, chat, and playgrounds you already use.
Memory operations are nodes on the workflow canvas. One node reads context, the next writes results.
Learn moreAny chat thread can reference a memory folder so the model has up-to-date context.
Learn morePrompt variables can take query results as input, so each run uses current data.
Learn moreMemory works alongside the prompts, workflows, and chat threads you already build. Vector search and external MCP access are included.