Products
Seren Memory
Give agents durable, user-owned private memory while organizational knowledge stays governed. Seren Memory separates personal recall from shared knowledge. Each domain keeps its own retrieval, review, and authorization rules.
Private memory
Recall private context across sessions
Private memory is restricted to the authenticated user and organization. Employees can recall decisions, preferences, procedures, verified fixes, and connected context across sessions.
Recall searches by meaning, exact words, time, and relationships. If semantic search is unavailable, recall uses the available retrieval methods.
Completed conversations can retain their original transcript as a private source while extracting only durable facts into recall. The source and recalled memories stay linked for provenance, but raw transcript text never enters normal semantic, keyword, or graph recall.
Seren uses TLS and encrypted database storage for Memory data. Private facts remain queryable inside the isolated organization database so recall can use meaning, words, time, and relationships. Keep credentials in Seren Passwords instead of memory content.
Agent lifecycle hooks
Bring relevant memory into each session
The Seren CLI connects Codex and Claude Code sessions to the same private memory. At session start, a hook injects a bounded set of reviewed memories as quoted reference material.
After each completed turn, another hook sends policy-permitted prompts and responses for structured fact extraction. A reviewed memory can inform later sessions on either platform.
Organization policy controls capture, supported platforms, prompt and response capture, transcript size, and cached-policy grace during control-plane outages. Automatic hooks do not retain the raw transcript payload in Seren Memory. A stable source ID prevents duplicate capture when the same path repeats a delivery.
A normalized Git remote or stable local mapping keeps workspace identity consistent across both agent platforms.
The CLI filters common secret patterns before delivery. Pattern filters cannot detect every secret or paraphrase. Tool output is not captured as a separate event. Text quoted into a prompt or response follows the policy for that field. Before delivery, the CLI applies the current policy again. A stale policy revision cannot silently deliver content under earlier rules.
Captured facts remain private to the authenticated user inside the organization. They do not enter a shared organizational knowledge domain. If Memory is unavailable, the CLI queues encrypted content when a secure local key is available. Without a secure key, it stores only metadata that requires attention. The agent session continues when capture cannot proceed.
seren memory hook status shows content-free queue counts, queue age, delivery outcomes, and policy rejections. It never prints captured prompts or responses.
Reviewable memory
Keep facts, drafts, and history distinct
Newly extracted facts start as unreviewed. Automatic session bootstrap injects only reviewed memories. The model-facing MCP tools cannot review their own memories. Lifecycle status separately marks content as active, draft, canonical, or deprecated.
Source links keep provenance visible. Time-aware relationships record when each connection was valid. Applications can inspect that relationship timeline or reconstruct the graph at an earlier time. The timeline describes relationship history, not complete content revision history.
claude-mem migration
Move existing memory through a controlled cutover
An opt-in CLI build imports claude-mem observations and summaries from consistent, read-only SQLite snapshots. It supports content-free inspection, reviewable plans, local rehearsals, repeatable imports, resume, verification, and provenance-scoped rollback. Standard Seren CLI and MCP builds do not include SQLite.
Imports preserve observed time, stable source identity, workspace mapping, and available claude-mem provenance. Imported facts start unreviewed. A migration cannot complete until all selected records are reconciled and their embeddings are ready. Rollback removes only records attributed to the selected run or migration series.
Regular snapshots can run while claude-mem remains active and Seren hooks remain disabled. During final cutover, stop claude-mem before the final catch-up. Verify that migration. Then enable Seren hooks. Do not run both automatic capture systems at the same time. Their different source identities cannot deduplicate the same turn.
Keep the source database and migration report as a read-only archive. Seren does not synchronize post-cutover memories back into claude-mem.
View the migration commandsPortability
Export private memory with its sources
Authenticated users can export their private memories together with retained conversation sources. Export pages keep the two collections separate, preserve source provenance, and remain scoped to the current user and organization. Applications continue through the shared page offset until no next offset is returned.
Connected sources
Bring your notes into memory
Bring in notes from Seren Notes or tools such as Notion. A synchronization run updates Memory when the source changes. The history and source remain visible, and the connected source remains the editing location.
Memory and working data
Use the right shared layer for the job
Memory helps employees recall meaning, context, decisions, and organizational knowledge. An organization-scoped skill database is the system of record for shared tasks, records, and transactions. Teams can use both layers together because each layer has a separate purpose.
Explore skill databasesEmployee memory configuration
Enable built-in memory with supported modes
A managed employee can enable built-in semantic and graph memory in its deployment configuration. That built-in memory is separate from Seren Memory and holds working context for the employee runtime.
Enabled semantic memory currently supports the Seren-managed store, tool-based reads, and writes that are either tool-based or disabled. Seren rejects an external store, always-on reads, and on-observation writes when you save the configuration, so an unsupported setting fails immediately instead of when the employee next starts.
Built-in semantic and graph memory tools are unavailable during an organization run. These tools do not support an organization-scoped runtime namespace. Use Seren Memory for organization knowledge.
Organizational knowledge
Shared context without open-ended access
Organizations can separate governed knowledge into general, financial, legal, and engineering domains. A domain can be organization-visible or restricted through direct user and agent grants.
Restricted agent access requires grants for both the authenticated user and the agent. A grant for one identity does not grant access to the other identity. Each domain keeps separate knowledge. Labels help people organize domains but never grant access.
People and agents can search only the domains that they can access. A domain search cannot return knowledge from another domain or organization.
Owners control domain access, editors maintain content, and agents have read access. Organization owners and administrators create domains. Seren records each access change.
An organization employee run refers to its selected knowledge through an opaque, provider-owned selection. Seren Memory maps that selection to the current domain and checks the user, employee, organization, and domain grant again when knowledge is read. The shared work context does not replace Memory's domain permissions or turn a domain into a general container for credentials, files, skills, or conversations.
Organization-targeted employee runs currently use only organization-visible domains. Seren rejects restricted domains for this output audience, even when the user and employee have direct grants. Restricted domains remain available through Memory operations that support their narrower audience.
Reading knowledge, capturing new memory, and choosing who receives output are separate decisions. Organization collaboration does not write the run into the selected domain. It also does not copy channel or organization content into personal memory.
There is no publishing step. Updates become searchable shortly after editors save them.
Seren uses TLS and encrypted database storage for Memory data. Each organization's knowledge stays separate, and Seren checks access on every request. Encryption does not replace user and agent grants.
Configure organization collaborationFor developers
Add memory to agent workflows
Use Seren Memory when an employee or application needs private recall, retained conversation provenance, point-in-time relationship history, governed organizational search, or a portable export. The TypeScript, Python, and Rust SDKs expose the same core memory and knowledge operations. The Seren CLI provides lifecycle hooks and the opt-in migration workflow. The API reference documents the exact service contracts.