Durable memory
Agents forget between sessions. A Memava vault does not. What one session learns, the next session reads.
Memava speaks MCP, the open Model Context Protocol. If your agent, IDE, or harness can call MCP tools, it can read and write a Memava vault as a named, permissioned team member. No SDK, no plugin, no lock-in to any one AI vendor. Ready-made setup guides for the popular tools live on the Integrations page; this page is the contract underneath them.
Claude Code takes one command; Codex takes the block below in ~/.codex/config.toml. Connect each tool as its own agent.
“What changed in the vault this week, and who did it?” It answers from the history, with names.
claude mcp add --transport http memava https://memava.com/o/YOUR-TEAM/mcp/YOUR-AGENT-TOKEN [mcp_servers.memava] url = "https://memava.com/o/YOUR-TEAM/mcp/YOUR-AGENT-TOKEN"
Every agent gets its own token and its own identity in the vault. Point any MCP client at the endpoint and the twenty-three tools are live:
POST https://memava.com/o/<team>/mcp/<agent-token> note_list what the agent is allowed to see note_read read one note note_search full-text search inside its grants (path: author: since: tag: has:task) note_query filter and sort notes by typed frontmatter properties note_history who changed a note, and when note_edit replace one exact passage, safe beside live human edits note_backlinks what links TO a note (the graph's other half) note_activity what teammates changed lately (catch-up in one call) note_create new note, inside an edit grant note_update replace or append, merged LIVE into open editors note_remember store a fact or inference with its source, as a plain note line note_recall recall remembered facts across readable notes, stale flagged at 90 days note_propose propose a change to any readable note; a reviewer accepts hunk by hunk task_list every task across readable notes: status, assignee, due, overdue flagged task_update set a task's status, assignee or due date, one live line, attributed note_skills the team's playbooks under skills/, scoped by the agent's grants note_batch many operations as one unit, dry run first, one id to undo note_batch_revert put every note a batch touched back, as new attributed commits note_related related notes with reasons, unlinked mentions, block references note_suggest propose tags, properties or a folder as one reviewable proposal note_inbox lines addressed to the agent (@atlas ...) across readable notes note_comment a comment or reply in a thread beside the note, attributed note_remind a reminder line on a note, a tap when it comes due
Writes apply as live CRDT operations: a person watching the note sees the agent's words merge in as they land, attributed to the agent in the note's history.
Reads and writes are gated by the same access rules that govern people, evaluated on every call, and the audit records what an agent read, not just what it wrote.
Agents forget between sessions. A Memava vault does not. What one session learns, the next session reads.
A research agent, a drafting agent, and a human editor can work the same vault, each under its own grants, each attributed.
The memory is plain markdown a person can open, audit, correct, and revert. No embeddings-only black box.
No delete tool exists. Private trees are unreachable. A leaked token is one revocation away from dead, and everything it did is in the history.
The same tools answer a REST door, for scripts and CI. Memava never reads your repositories: your pipeline writes what merged, and every agent on the team reads it.
# Once: create the log note.
curl -s https://memava.com/o/YOUR-TEAM/mcp-rest/note_create \
-H "Authorization: Bearer $MEMAVA_TOKEN" -H 'content-type: application/json' \
-d '{"path":"log/merges.md","body":"# Merges\n"}'
# On every merge: append a line, and fail the job if Memava refused it.
curl -s https://memava.com/o/YOUR-TEAM/mcp-rest/note_update \
-H "Authorization: Bearer $MEMAVA_TOKEN" -H 'content-type: application/json' \
-d '{"path":"log/merges.md","body":"- Merged: the matching feature (PR 412)\n","mode":"append"}' \
| grep -q '"ok":true'
In Access, give the CI agent an edit grant on log/ and remove the whole-vault grant every new agent starts with. Every line it writes carries its name in the history.
Agencies and independent builders run agents for their clients, and every one of those clients needs a memory their agent can use and their team can see. Memava is built for exactly that shape: each client gets a fully isolated vault, you get an operator's setup path, and your client gets a product they can open, read, and trust.
There is a real program for this: partner terms, a walkthrough of the operator side, and revenue share for the teams you bring. The partner page has the numbers.