NEXO Brain is a cognitive AI memory and orchestration server that transforms stateless AI agents into persistent, learning partners with 150+ MCP tools.
Cognitive Memory & RAG
Semantic retrieval over Short-Term (STM) and Long-Term (LTM) memory stores
Memory metrics, inspection by ID, Ebbinghaus forgetting curves, and rehearsal-based reinforcement
Cognitive dissonance detection and resolution
User sentiment detection, tone guidance, and adaptive trust scoring (0–100)
Session Lifecycle & Multi-Session Coordination
Register, heartbeat, and cleanly close sessions; list active sessions
Track files across sessions to detect edit conflicts
Inter-session messaging: fire-and-forget, broadcast, ask/answer exchanges
Episodic Memory (Changes, Decisions, Diary)
Log code/config changes with full context; log decisions with reasoning and outcomes
Write/read end-of-session diaries for continuity
Unified recall search across all memory types simultaneously
Learnings & Guard System
Add, search, update, and delete learned patterns, error gotchas, and operational knowledge
Pre-action guard checks surface relevant learnings before code edits to prevent repeated mistakes
Tasks, Reminders & Followups
Create, update, complete, and delete user reminders and autonomous followup tasks (with recurrence)
Log operational task executions and detect overdue tasks
Knowledge & Entity Management
Store and manage entities (people, services, URLs) and user preferences by category
Securely store, retrieve, and manage credentials (API keys, tokens) locally in SQLite
Agent Registry
Register, update, and manage AI agent profiles with specialized models, rules, and toolsets
Evolution & Self-Improvement
Weekly self-improvement cycles with proposal, approval, and rejection mechanisms
View evolution dimension scores and history
Data & Search Management
Database backup, list, and restore
FTS5 full-text search indexing for directories and files with dynamic directory registration
Extensibility & Operations
Hot-loadable plugin system with base + personal override directories
Unified operations menu (nexo_menu) with alerts and active session overview
Scans and analyzes local Git repositories to learn project context, development patterns, and history, incorporating this data into the assistant's long-term memory.
This repository contains the local runtime core that powers NEXO Desktop.
NEXO Desktop is the public product and the supported user-facing installer,
update channel, onboarding surface, settings UI, and support path. The legacy
nexo-brain package and binary names remain temporarily for install/update
compatibility only; they are not a separate product line.
Version 7.38.7 is the current packaged-runtime line
Public product: NEXO Desktop
Runtime role: local NEXO core bundled with Desktop
Active systems: local memory, Deep Sleep, Evolution support-ticket mode, Skills, Watchdog, followups, doctor diagnostics, and MCP tooling
Evolution mode: enabled by default for Desktop-managed installs; it never opens GitHub branches, pushes, PRs, transcripts, local databases, or raw private evidence, and routes product-improvement requests through sanitized support tickets
Distribution: Desktop installers and update manifests are published through the NEXO Desktop release channel
Use nexo-desktop.com for downloads, billing, support, and product information.
Related MCP server: mcp-memory
Release changes that touch bootstrap, session handling, Deep Sleep, client
sync, or shared runtime claims must also keep docs/client-parity-checklist.md
current.
Managed via bootstrap + Codex config mcp_servers.nexo
Managed MCP-only shared-brain metadata
Runtime doctor parity audit
Answer a pending question from another session.
Inputs: qid (required), answer (required).
Make an authenticated HTTP request to the NEXO Desktop backend (nexo-desktop.com).
The session bearer is auto-loaded from the OS keychain — the agent never sees or handles tokens. Use this for any /api/* endpoint the user has permission for: provider-proxy/, credits/, cards/*, auth/app-tokens, etc.
Inputs: path (required), method (required), base_url, body_json, +2 more.
Ask a question to another session (they see it on next heartbeat).
Inputs: to_sid (required), from_sid (required), question (required).
Build or apply the safe automation reconciliation plan.
Inputs: apply.
Read-only supervisor report for automations, cron runs, cron spool and Evolution policy.
Inputs: markdown.
Explain one NEXO product capability with source and safety context.
Inputs: query, locale, capability_id.
Find official NEXO protocol cards for a user request.
Inputs: query (required), limit, locale, category, +2 more.
Check if a question has been answered.
Inputs: qid (required).
Read the latest session checkpoint. Used by PostCompact hook and for manual recovery.
Inputs: sid.
Save a session checkpoint for auto-compaction continuity.
Call this BEFORE context compaction to preserve session state. The PostCompact hook reads this checkpoint and re-injects it as a Core Memory Block, so the session continues seamlessly.
Inputs: sid (required), task, next_step, task_status, +5 more.
Close a verified closure item or reject/stale it explicitly.
Inputs: item_id (required), reason.
Return one closure item with sources and events.
Inputs: item_id (required).
Link a closure item to a task, workflow, followup, outcome, or learning.
Inputs: item_id (required), link_id (required), link_type (required), relation.
Return the next ranked closure items without executing source actions.
Inputs: area, kind, limit, state, +3 more.
Write and return an Operational Closure Plane daily snapshot.
Inputs: limit, refresh, snapshot_date.
Read the Operational Closure Plane status and ranked closure queue.
Inputs: limit, refresh.
Triage a closure item without executing its source action.
Inputs: item_id (required), kind, owner, state, +6 more.
Record verification evidence for a closure item.
Inputs: item_id (required), evidence (required).
Read-only metrics for Local Context, learnings, followups and intraday memory.
Inputs: window_seconds.
Decide whether an answer should proceed directly or be verified first.
Inputs: goal (required), area, stakes, unknowns, +5 more.
Build a context packet for subagent injection. Returns learnings + changes + followups + preferences + cognitive memories for a specific area.
MUST call before delegating ANY task to a subagent. Inject the result into the subagent's prompt.
Inputs: area (required), files.
Return compact local context evidence suitable for injection before a reply.
Inputs: query (required), limit, intent, max_chars, +1 more.
Return the forensic continuity timeline for a conversation.
Inputs: conversation_id (required), limit.
Persist a compaction-related continuity event into the canonical snapshot stream.
Inputs: conversation_id (required), payload, trace_id, event_type, +1 more.
Build the small continuity bundle Desktop injects after restore or stale resume loss.
Inputs: client, session_id, token_budget, conversation_id, +1 more.
Read recent continuity snapshots by conversation_id or session_id.
Inputs: limit, session_id, conversation_id.
Write a durable continuity snapshot for Desktop/Brain handoff.
Inputs: conversation_id (required), client, payload, trace_id, +4 more.
Cognitive pre-action check. Call before significant actions.
Inputs: goal (required), plan, unknowns, task_type, +4 more.
Evaluate concrete alternatives for a high-impact task and persist the recommendation.
Inputs: goal (required), alternatives (required), area, goal_id, +10 more.
Create a persistent AppToken for the current user via POST /api/auth/app-tokens.
Use this when a card needs to mint a token that will live inside a snippet the user pastes on their own website (chatbot widget, embed, public API autoresponder). The plain-text token is returned ONCE — embed it in the generated snippet and never store it elsewhere.
Inputs: name (required), abilities, expires_at, allowed_platforms.
Store a new credential.
Inputs: key (required), value (required), service (required), notes.
Delete credential(s). If no key, deletes all for the service.
Inputs: service (required), key.
Get credential value(s) for a service.
Inputs: service (required), key.
List credentials (names and notes only, no values).
Inputs: service.
Update a credential's value and/or notes.
Inputs: key (required), service (required), notes, value.
Mark a drive signal as investigated with an outcome.
Call this after NEXO has autonomously investigated a READY signal.
Inputs: outcome (required), signal_id (required).
Dismiss a drive signal (archived, not deleted).
Call this when a signal is not worth investigating.
Inputs: reason (required), signal_id (required).
Reinforce a drive signal with a new observation.
Increases tension and may promote the signal status (latent → rising → ready).
Inputs: signal_id (required), observation (required).
List autonomous drive/curiosity signals.
Drive signals are observations NEXO accumulates during normal work. When tension crosses threshold, NEXO investigates silently.
Inputs: area, limit, status.
Return read-only status for the cognitive embedding migration.
Inputs: no named parameters.
Build a full local dossier for one entity with aggregates and evidence.
Inputs: query (required), max_chars, max_facts, max_assets, +1 more.
Record a compact evidence pointer without storing raw command output.
Inputs: summary (required), action, source_id, object_ref, +6 more.
Search concrete evidence across tasks, workflows, change logs, diaries and continuity stores.
Inputs: limit, query, task_id, artifact, +3 more.
Show all tracked files across all active sessions with conflict detection.
Inputs: no named parameters.
Mark a followup as completed. Appends result to verification field.
Inputs: id (required), result.
Create a new agent followup (autonomous task).
Inputs: id (required), description (required), date, owner, +6 more.
Soft-delete a followup.
IMPORTANT: call nexo_followup_get first and pass its READ_TOKEN.
Inputs: id (required), read_token.
Read a followup with its history and usage rules.
IMPORTANT: before update/delete/restore/note, call this tool first and use the returned READ_TOKEN.
Inputs: id (required).
Return followups grouped by lifecycle lane for runner, dashboard and startup parity.
Inputs: limit.
Append a note to followup history.
IMPORTANT: call nexo_followup_get first and pass its READ_TOKEN.
Inputs: id (required), note (required), actor, read_token.
Restore a soft-deleted followup back to PENDING.
IMPORTANT: call nexo_followup_get first and pass its READ_TOKEN.
Inputs: id (required), read_token.
Update fields of an existing followup. Only non-empty fields are changed.
IMPORTANT: call nexo_followup_get first and pass its READ_TOKEN.
Inputs: id (required), date, owner, status, +5 more.
Read one durable goal and optionally include linked workflow runs.
Inputs: goal_id (required), include_runs.
List durable goals.
Inputs: limit, status, include_closed.
Open a durable goal so objectives survive sessions.
Inputs: sid (required), title (required), owner, priority, +5 more.
Update a durable goal with active/blocked/completed state.
Inputs: goal_id (required), owner, title, status, +6 more.
Check learnings relevant to files/area before reading or editing code.
Inputs: area, files.
Temporarily override a Guardian rule's mode (Plan Consolidado 0.17).
Writes to ~/.nexo/config/guardian-runtime-overrides.json which
guardian_config.rule_mode already honours at read time. Useful when
a rule is noisy during an incident and needs to drop to shadow for an
hour without a server restart.
Inputs: rule_id (required), ttl, mode.
Update session task, check inbox and pending questions. Auto-detects trust events.
Call this at the START of every user interaction (before doing work). May surface silent runtime signals DIARY_OVERDUE, GUARD_REMINDER, LEARNING_REMINDER, and PROTOCOL_DEBT; clients must treat those as internal obligations, not user-visible content. Output always begins with a NOW_UTC line (ISO-8601, UTC) — use it as the authoritative wall-clock time for any artifact (emails, diaries, followups) to avoid date/day-of-week drift across long sessions. Args: sid: Your session ID from nexo_startup. task: Brief description of current work (5-10 words). context_hint: Last 2-3 sentences from the user or current topic. Used for sentiment detection, trust auto-scoring, and mid-session RAG. ALWAYS provide this for best results.
Inputs: sid (required), task (required), context_hint.
List recent hook lifecycle runs and per-hook health summary.
Closes Fase 3 item 7 of NEXO-AUDIT-2026-04-11. Each NEXO hook (session-start, post-compact, pre-compact, inbox-hook, etc.) writes a row to hook_runs when it finishes via scripts/nexo-hook-record.py. This tool reads them back so the agent can answer "is the hook pipeline healthy?" without needing the dashboard or grepping log files.
Inputs: hours, limit, status, hook_name, +1 more.
List hot-context items currently alive in the recent continuity window.
Inputs: hours, limit, state.
Register a new directory for FTS5 search indexing. Survives restarts.
Inputs: path (required), notes, dir_type, patterns.
List all directories being indexed by FTS5 (builtin + dynamic).
Inputs: no named parameters.
Remove a directory from FTS5 indexing and clean up its entries.
Inputs: path (required).
Run a low-limit daytime memory observation cycle that only publishes evidence-backed intraday facts.
Inputs: limit, backfill_limit, pending_sla_seconds.
Add a new learning (resolved error, pattern, gotcha).
Inputs: title (required), content (required), category (required), priority, +6 more.
Scan recent decisions and surface those that conflict with a learning's prevention rule.
Closes Fase 2 item 3 of NEXO-AUDIT-2026-04-11. Use this when you add a new
rule and want to retroactively check whether past decisions still hold.
Creates deterministic NF-RETRO-L-D followups so the
helper is idempotent across reruns. nexo_learning_add invokes this
automatically when the new learning has a prevention field — call this
tool manually only when you want to re-scan with a longer window or a
different threshold.
Inputs: learning_id (required), dry_run, min_score, max_matches, +1 more.
Delete a learning entry.
Inputs: id (required).
List all learnings, grouped by category.
Inputs: limit, category, created_after, created_before.
Score learning quality so fragile rules can be strengthened before they mislead guard or retrieval.
Inputs: id, limit, status, category.
Dry-run the canonical learning resolver without creating or updating learnings.
Inputs: title (required), content (required), category (required), priority, +5 more.
Search learnings by keyword. Searches title and content.
Inputs: query (required), limit, category.
Update a learning entry. Only non-empty fields are changed.
Inputs: id (required), title, status, content, +7 more.
Return one indexed local asset by asset id.
Inputs: asset_id (required).
Return graph relations around one indexed local asset.
Inputs: asset_id (required), limit.
Retrieve local evidence before answering or acting.
Use mode='compact' for normal answers. Use mode='full' only for deep debugging, ideally with a higher max_chars and a specific query.
Inputs: query (required), mode, limit, intent, +5 more.
Control the local memory index.
Inputs: root, limit, action, process_limit, +1 more.
Return recent local memory diagnostic log entries.
Inputs: limit.
List, add or remove local memory exclusions.
Inputs: path, action, reason.
List, include, exclude or reset local memory file extension rules.
Inputs: mode, action, reason, extension.
Plan or apply Local Memory roots v2 cleanup.
Inputs: apply.
Return or warm local model state for the Local Context Layer.
Inputs: action, local_files_only.
Purge one indexed asset or clear the full local index.
Inputs: asset_id, clear_all.
List, add or remove local memory roots.
Inputs: mode, path, depth, action.
Render service configuration metadata for macOS, Windows or Linux.
Inputs: platform_name.
Return local memory index status for Desktop settings and support diagnostics.
Inputs: no named parameters.
Read or apply the Brain-owned managed MCP catalog reconciliation plan.
Inputs: apply.
Return durable MCP write queue counts and recent records.
Inputs: limit.
Answer a memory question only when evidence exists.
Inputs: query (required), limit, time_range, project_hint.
Backfill Memory Observations v2 from existing Brain tables.
Inputs: limit, sources.
List raw Memory Observations v2 events captured by hooks/tasks.
Inputs: limit, query, source_id, event_type, +3 more.
Summarize raw Memory Observations v2 event counts.
Inputs: days.
SELECTIVE-FORGET: verifiably erase a revoked secret, or correct a fact.
Two modes, never mixed:
mode='secret' (HARD-FORGET): scrubs the secret from EVERY table of ALL LIVE DBs the agent retrieves from — nexo, cognitive, local-context, local-context-usage, and email — discovered by live introspection (sqlite_master + PRAGMA) over each subsystem's own canonical path resolver, not a curated list — plus every FTS index (incl. local_chunks_fts), on-disk transcripts, and legacy shadow DBs (cognitive
local-context). The row is DELETED where it IS the secret (or carries an embedding/FTS copy), and REDACTED in place where the secret is embedded in an otherwise-useful record (diary, item_history.note, local_chunks.text, entity_facts.value, change_log...). It then RE-ENUMERATES and RE-SCANS everything and only returns complete=True at total zero; any survivor (even in a table not anticipated) is reported as complete=False with its . location. Use ONLY for revoked credentials/secrets or toxic data. Destructive run requires dry_run=False AND confirm=True (a dry-run count is returned otherwise). HNSW persisted indices are invalidated. SCOPE: point-in-time backups / snapshots are NOT swept (retention) — rotate the secret; the report's backup_scope field states this so complete=True never overclaims.
mode='fact' (CORRECT-FACT): does NOT delete. Useful memory is preserved via the existing reversible supersede; correct it with nexo_learning_update / supersede instead.
Inputs: value (required), mode, reason, confirm, +2 more.
Return Memory Observations v2 health and table status.
Inputs: no named parameters.
Run safe Memory Observations v2 maintenance.
Inputs: retry_failed, process_limit, backfill_limit, backfill_sources.
List passive Memory Observations v2 rows.
Inputs: limit, query, status, session_id, +2 more.
Process pending raw memory events into passive observations.
Inputs: limit, backfill_limit, pending_sla_seconds.
Summarize passive Memory Observations v2 rows and queue status.
Inputs: days.
Search Memory Observations v2 with evidence-first results.
Inputs: query (required), depth, limit, time_range, +1 more.
Return a chronological Memory Observations v2 timeline.
Inputs: limit, query, time_range, project_hint.
Record proposal feedback and apply suppression/snooze when requested.
Inputs: feedback (required), proposal_id (required), note, snooze_until.
Return one opportunity with evidence and read-only preparations.
Inputs: opportunity_id (required), include_evidence.
Return at most three evidence-backed proposals for a user surface.
Inputs: limit, refresh, surface, include_snoozed.
Generate Opportunity Orchestrator candidates from existing evidence.
Inputs: dry_run, sources, write_report, limit_per_source.
Suppress repeated opportunity suggestions by scope.
Inputs: scope_key (required), scope_type (required), reason, expires_at.
List all loaded plugins and their tools, showing source (repo/personal).
Inputs: no named parameters.
Load or reload a plugin. Searches repo plugins/ first, then NEXO_HOME/personal/plugins/.
Inputs: filename (required).
Unregister a plugin's tools from MCP (does not delete files).
Inputs: filename (required).
Build the 24h recent-context bundle that should be reviewed before acting.
Especially useful for emails, orchestrators, and any work where the same topic may reappear hours later.
Inputs: hours, limit, query, session_id, +1 more.
Route a user turn through pre-answer continuity sources before responding.
Inputs: sid, area, files, query, +7 more.
Answer a NEXO product question using the structured product catalog.
Inputs: question (required), limit, locale.
Search the structured NEXO product capability catalog.
Inputs: limit, query, status, category.
Validate the structured NEXO product knowledge catalog.
Inputs: no named parameters.
Show which NEXO product capabilities are exposed by a product surface.
Inputs: limit, surface.
Resolve protocol debt records by id or filters.
Inputs: debt_id, task_id, debt_ids, debt_types, +2 more.
Read recent hot context and continuity events from the last N hours.
Inputs: hours, limit, query, context_key.
Capture/update a recent 24h context item and append an event.
Use this for important ongoing threads that should stay mentally fresh across sessions/clients.
Inputs: title (required), actor, owner, state, +9 more.
Resolve a recent hot-context item and append a resolution event.
Inputs: actor, topic, source_id, ttl_hours, +4 more.
Force full rebuild of the FTS5 search index. Use after bulk changes or if search seems stale.
Inputs: no named parameters.
Mark a reminder as completed with today's date.
Inputs: id (required).
Create a new reminder for the user.
Inputs: id (required), description (required), date, owner, +2 more.
Soft-delete a reminder.
IMPORTANT: call nexo_reminder_get first and pass its READ_TOKEN.
Inputs: id (required), read_token.
Read a reminder with its history and usage rules.
IMPORTANT: before update/delete/restore/note, call this tool first and use the returned READ_TOKEN.
Inputs: id (required).
Append a note to reminder history.
IMPORTANT: call nexo_reminder_get first and pass its READ_TOKEN.
Inputs: id (required), note (required), actor, read_token.
Restore a soft-deleted reminder back to PENDING.
IMPORTANT: call nexo_reminder_get first and pass its READ_TOKEN.
Inputs: id (required), read_token.
Check reminders and followups.
Inputs: filter.
Update fields of an existing reminder. Only non-empty fields are changed.
IMPORTANT: call nexo_reminder_get first and pass its READ_TOKEN.
Inputs: id (required), date, owner, status, +4 more.
Return the resident NEXO Runtime Service status for diagnostics.
Inputs: no named parameters.
Audit stores that write data without verified later consumption.
Inputs: markdown.
Send a fire-and-forget message to another session or broadcast.
Inputs: text (required), to_sid (required), from_sid (required).
Return Brain-verifiable heartbeat, diary, learning, and close compliance state.
Inputs: sid, diary_window_minutes.
Read recent session diaries for context continuity.
Inputs: brief, domain, last_n, last_day, +1 more.
Write end-of-session diary with decisions, pending context, and self-critique.
Exposed as an essential MCP tool because Desktop close/archive/app-exit enforcement depends on it even when dynamic plugin loading is disabled.
Inputs: domain, source, pending, summary, +8 more.
Export a machine-readable session bundle for cross-client handoff or archival.
Inputs: sid, path.
Close an automation_runs row opened by nexo_session_log_create.
Inputs: session_id (required), error, returncode, cost_source, +6 more.
Open an automation_runs row for an interactive Claude/Codex session.
Designed for clients that spawn Claude/Codex directly (notably NEXO Desktop, which runs a TypeScript process that shells out to the CLI without going through run_automation_prompt). Call this BEFORE spawning the child, store the returned session_id, then call nexo_session_log_close when the session ends.
Inputs: caller (required), backend (required), cwd, pid, +5 more.
Build a portable handoff packet for another client/runtime.
Use this when another client should continue the same work with explicit task/checkpoint/goal/workflow context instead of relying on memory alone.
Inputs: sid.
Find reusable NEXO skills for the current task.
Inputs: task (required), level.
Pre-load relevant cognitive memories based on pending followups, due reminders, and last session topics.
Call during startup (after nexo_startup) to ensure the session starts with the right context loaded. Returns up to 10 memories matching the current operational state.
Inputs: no named parameters.
Register new session, clean stale ones, return active sessions + alerts.
Call this ONCE at the start of every conversation. Returns the session ID (SID) — store it for use in all other nexo_ tools.
Inputs: task, session_token, session_client, conversation_id, +2 more.
List active sessions. Filter by keyword if provided.
Inputs: keyword.
Cleanly close a session. Removes it from active sessions immediately.
Call this when ending a conversation to avoid ghost sessions. Args: sid: Session ID to close.
Inputs: sid (required).
Close a real NEXO support ticket after evidence has been recorded.
Inputs: ticket_id (required).
Create a real NEXO support ticket for a product bug/setup issue.
Inputs: message (required), subject (required), origin, priority, +1 more.
List real support tickets from the NEXO backend for the signed-in Desktop user.
Use this when the user asks about support tickets or bug reports. Do not substitute a private followup for an actual product support ticket.
Inputs: limit, status.
Append an evidence note to a real NEXO support ticket before status changes.
Inputs: body (required), ticket_id (required), client_message_id.
Read one real support ticket from the NEXO backend by id.
Inputs: ticket_id (required).
Reopen a real NEXO support ticket if fresh evidence shows it is active.
Inputs: ticket_id (required).
Read NEXO's live system catalog built from core tools, plugins, skills, scripts, crons, projects, and artifacts.
Inputs: limit, query, section.
Acknowledge blocking guard rules on an open protocol task.
Inputs: sid (required), task_id (required), note, learning_ids.
Close a protocol task with evidence and optional artifacts.
For high-stakes/irreversible closures (publish stable, broadcast, payment,
force-push, revoke) pass work_type/stakes plus artifact_hash and
last_human_validation_of_artifact_hash (both must match) so the close
gates can be satisfied instead of dead-ending.
Inputs: sid (required), task_id (required), result, stakes, +29 more.
Check which operational tasks are overdue based on their frequency.
Compares last execution date vs configured frequency. Returns overdue tasks or 'all tasks up to date'.
Inputs: no named parameters.
Show execution history for operational tasks.
Inputs: days, task_num.
Record that an operational task was executed.
Inputs: task_num (required), task_name (required), notes, reasoning.
Open a protocol task for non-trivial work.
ack_rules accepts "#95,#156" / "95,156" / "[95, 156]" and, when
the guard surfaces blocking rules, acknowledges them inline instead
of requiring a separate nexo_task_acknowledge_guard call.
Inputs: sid (required), area, goal, plan, +12 more.
Explain a live NEXO tool/capability from the generated system catalog.
Inputs: name (required).
Track files being edited. Detects conflicts with other sessions.
MUST call before editing any shared file. Args: sid: Your session ID. paths: List of absolute file paths to track.
Inputs: sid (required), paths (required).
Read a full transcript fallback by session id, transcript display name, session_uid, or exact path.
Inputs: client, session_ref, max_messages, transcript_path.
List recent Claude Code / Codex transcripts visible to NEXO.
Inputs: hours, limit, client.
Search recent transcripts directly when recall/hot-context are not enough.
Inputs: hours, limit, query, client.
Stop tracking files. If no paths given, releases all.
Inputs: sid (required), paths.
Return the compensation plan for a partially completed workflow.
Inputs: run_id (required), checkpoint_limit.
Read the full durable workflow state.
Inputs: run_id (required), include_steps, checkpoint_limit.
Record a durable workflow handoff.
Inputs: actor (required), run_id (required), new_owner, next_action, +2 more.
List durable workflow runs.
Inputs: limit, status, include_closed.
Open a durable workflow run for long multi-step work.
Inputs: sid (required), goal (required), owner, steps, +7 more.
Replay recent checkpoints for a workflow run.
Inputs: run_id (required), limit.
Summarize the next actionable step for a workflow run.
Inputs: run_id (required).
Update a workflow run with a replayable checkpoint.
Inputs: run_id (required), actor, owner, summary, +14 more.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 170 tools
Every tool has a distinct resource+verb (get/update/delete/list by entity), but with 170 overlapping entities there is inevitably semantic overlap (e.g. multiple 'explain' or 'surface' tools); still, each appears to target a unique function.
Uniform nexo_ prefix with snake_case verbs and nouns is extremely consistent across all 170 tools; pattern is predictable.
170 tools is an extreme surface area. Even intentional 'mission control' servers rarely need this many; it overwhelms context and decision-making.
For the apparent domain (multi-session memory, checkpoints, events, sessions, products, external APIs), lifecycle coverage is thorough — create/update/close/query pairs exist across the board. Some areas (e.g., plugin removal edge cases) may still have gaps, hence not 5.
Persistent memory and knowledge management for AI agents with semantic search and 50+ tools.
Memory, reflection and safety for AI agents. Hosted or local, no vector database.
Persistent memory, knowledge graph, task management and more for AI agents via MCP.
Universal memory for AI agents and tools. Save, organize and search context anywhere.
