Manage persistent AI memories organized by tier (core, active, archive) and category (preference, pattern, decision, correction, knowledge, general, user_profile, example).
| Tool | Description | Required Params |
|---|
memory_list | List memories. Filter by tier, category, or project. | — |
memory_search | Semantic search across memories. | query |
memory_load_context | Load core + relevant active memories. Pass mode='lean' to get id+title handles only (~70% smaller) then fetch bodies with memory_get. | — |
memory_get | Fetch full memory bodies by id. Pair with mode='lean' to load just what you need. | ids |
memory_create | Create a new memory. Use category='example' for canonical worked examples that ground the agent's behaviour. | title, content |
memory_update | Update an existing memory. | id |
memory_delete | Delete a memory by ID. | id |
memory_promote | Change tier (archive ↔ active ↔ core). | id, tier |
memory_stats | Memory usage statistics — count, tokens, categories. | — |
memory_image_upload | Upload an image to a memory with OCR. | memory_id, image_data, filename, mime_type |
memory_image_list | List images attached to a memory. | memory_id |
Track what AI discussed and accomplished across sessions. Auto-summarized and embedded for semantic search.
| Tool | Description | Required Params |
|---|
conversation_log | Log what you discussed or accomplished. Pass domain (coding|research|support|ops|general) so the summariser variant matches your agent. | summary |
conversation_history | Load recent conversation history. Pass mode='compact' for a deterministic statistical rollup of older entries + recent_window verbatim — keeps long-running agents under their context budget. | — |
conversation_search | Semantic search across conversation history. | query |
| Tool | Description | Required Params |
|---|
projects_list | List all projects with feature/entry counts. | — |
project_get | Get project details, features, tech stack, scan info. | project_slug |
project_create | Create a new project. | name, slug |
project_init_local | Create a lightweight project with no GitHub repo or scan required — start preserving context immediately. | name |
project_update | Update project name, description, or settings. | project_slug |
project_delete | Delete project and all data. Permanent. | project_slug, confirm |
| Tool | Description | Required Params |
|---|
context_save | Save a context entry for a feature. | project_slug, feature_name, content |
context_get | Retrieve context entries, optionally filtered by feature. | project_slug |
context_bundle | Full project context as markdown — memories, features, tech stack. | project_slug |
session_start | Unified session start. Full mode returns XML-structured output (<project>, <core_memories>, <examples>, <features>, <recent_activity>, <context_budget>). Pass mode='lean' for a compact handle manifest ~70% smaller. Omit project_slug to auto-detect. | — |
| Tool | Description | Required Params |
|---|
scan_trigger | Trigger a cloud scan. Returns scan ID for tracking. | project_slug |
scan_status | Check scan progress — percentage, logs, results. | scan_id |
diff_analyze | Analyze a git diff and save feature-level changes. | project_slug, diff |
scan_on_push | Toggle auto-scan on push via GitHub webhook. | project_slug, enabled |
| Tool | Description | Required Params |
|---|
feature_list | List all features for a project. | project_slug |
feature_get | Get detailed knowledge for a feature — decisions, gotchas, dependencies. | project_slug, feature_name |
feature_update | Update a feature name or description. | feature_id |
feature_delete | Delete a feature and its context entries. | feature_id |
Multi-phase project plans with IDE integration and status tracking.
| Tool | Description | Required Params |
|---|
plan_list | List active plans for a project. | project_slug |
plan_get | Get plan details with phases and messages. | plan_id |
plan_update_phase | Update a phase status. | phase_id, status |
plan_phases | Get active phases as markdown for .remb integration. | project_slug |
plan_create_phase | Add a new phase to a plan. | plan_id, title |
plan_complete | Mark a plan as completed. | plan_id |
Traverse the knowledge graph — entities, relationships, code dependencies.
| Tool | Description | Required Params |
|---|
graph_query | Multi-hop traversal from any entity (up to 4 hops). | entity_type, entity_id |
graph_related | Everything related to a feature — memories, context, conversations. | feature_id or project_slug + feature_name |
explore_code_graph | Browse code-level graph — files, functions, classes, imports. | project_slug |
search_code_symbols | Semantic search across code symbols. | query, project_slug |
| Tool | Description | Required Params |
|---|
cross_project_search | Search across ALL projects for features, context, memories. | query |
cross_project_patterns | Find architectural patterns shared across projects. | pattern_query |
| Tool | Description | Required Params |
|---|
preview_deploy | Deploy a live preview to Vercel. | project_slug |
preview_list | List previews for a project. | — |
preview_status | Check deploy status. | preview_id |
preview_stop | Stop an active preview. | preview_id |
preview_delete | Delete a preview record. | preview_id |
| Tool | Description | Required Params |
|---|
file_read | Read a file from the project's GitHub repo. | project_slug, path |
file_tree | Browse the file tree. | project_slug |
| Tool | Description | Required Params |
|---|
audit_list | List recent AI code audit reports. | project_slug |
audit_get | Get full audit details with findings. | audit_id |
audit_trigger | Trigger an on-demand code audit. | project_slug |
Multi-agent coordination — prevent conflicts when multiple AI agents work on the same project.
| Tool | Description | Required Params |
|---|
action_log | Record an action with reasoning. | action, intent, reasoning, targets |
action_query | Check what other agents did recently. | project_id |
action_check | Check if files had agent actions. | project_id, targets |
action_claim | Claim a scope so others avoid it. | project_id, scope, reasoning |
action_release | Release a claimed scope. | project_id, scope |
Track events, create checkpoints, and manage file history across AI sessions.
| Tool | Description | Required Params |
|---|
track_event | Record session events — context switches, edits, branch changes. | project_slug, event_type |
checkpoint_create | Snapshot files before bulk edits. | project_slug, file_paths |
checkpoint_restore | Restore files from a checkpoint. | snapshot_group |
checkpoint_list | List available checkpoints. | project_slug |
file_history | Version history of a file across scans. | project_slug, file_path |
file_restore | Get content of a specific file version. | snapshot_id |
session_timeline | Recent activity timeline. | — |
diff_between_scans | Compare files changed between two scans. | — |
Personal-memory primitives — current focus, blockers, lessons, recurring problems, people — plus proactive helpers that flag decision conflicts and synthesize a weekly brief.
| Tool | Description | Required Params |
|---|
focus_set | Set the user's current focus (singleton per scope, stored as core memory). | content |
focus_get | Get the current focus for a scope. | — |
blocker_add | Record something the user is blocked on (timestamps the blocker). | title |
blocker_resolve | Mark a blocker resolved; optionally promote to a lesson. | id |
blocker_list | List active blockers with days-since-blocked. | — |
lesson_add | Record a lesson learned. | title, content |
lesson_list | List lessons learned, newest first. | — |
recurring_problem_log | Log an occurrence; flags pattern_detected on the 3rd hit. | problem |
person_set | Add or update a person in the user's context (upsert by name). | name |
person_list | List people in the user's context. | — |
decision_check | Before recommending an action, surface prior decisions that may contradict it. | proposed_action |
weekly_brief | Structured brief: wins, decisions, blockers, recurring patterns, recommendation. | — |
Cross-project user profile — name, role, communication style, preferred stack, things to never do. Stored as global memories with category='user_profile' so they apply to every project automatically.
| Tool | Description | Required Params |
|---|
profile_set | Add or update a profile field (e.g. name, role, stack, never_do). | key, value |
profile_get | Read all profile fields, or one by key. | — |
profile_clear | Delete a profile field. | key |
Procedural memory — reusable skills the agent can search, load, and self-heal. Store HOW to do something (triggers, exact commands, gotchas, verification steps); auto-suggested at session start when the user's task semantically matches.
| Tool | Description | Required Params |
|---|
skill_create | Save a reusable procedural skill (markdown body up to 64 KB). | name, content |
skill_list | List the user's skills, optionally scoped to a project. | — |
skill_load | Return the full markdown content of a single skill. | asset_id |
skill_suggest | Semantic match: given a task description, return top-N skills to consider loading. | query |
skill_record_outcome | Record success/failure for a skill; flags skills that need review. | asset_id, success |
skill_patch | Self-healing: replace an exact substring of a skill's content. Creates a new version. | asset_id, old_text, new_text |
File-based context offload. Write large tool outputs (scans, audits, diffs) to a session-scoped scratchpad path so they stay out of the prompt; read on demand. 256 KiB cap per entry, Redis-backed with 24h TTL.
| Tool | Description | Required Params |
|---|
scratchpad_write | Save large content to a session-scoped path. Path must be relative; no '..'. | session_id, path, content |
scratchpad_read | Read content previously written. | session_id, path |
scratchpad_list | List all paths in the session scratchpad. | session_id |
scratchpad_delete | Delete an entry. | session_id, path |
Typed session-handoff payloads. At session end, save a briefing (focus, decisions, blockers, recent files, working-memory keys) so the next session loads it as a markdown prompt block in one tool call instead of replaying the full transcript. Each successful load silently extends the 30-day TTL so active sessions never expire.
| Tool | Description | Required Params |
|---|
briefing_save | Save typed briefing for a session. | session_id |
briefing_load | Load latest briefing as a rendered markdown block. | session_id |
Detect context poisoning. Score the downstream outcome of every memory you act on — success, rejection, or undo (weighted strongest). Quarantine candidates surface memories whose negative-outcome score exceeds the threshold.
| Tool | Description | Required Params |
|---|
memory_feedback | Record outcome (success | rejection | undo) for a retrieved memory. | memory_id, outcome |
memory_quarantine_candidates | List memories whose negative-outcome score ≥ threshold (default 0.7). | — |
Typed planner → researcher → implementer → reviewer contracts. Bad transitions or missing payload keys throw at the boundary instead of silently downstream.
| Tool | Description | Required Params |
|---|
role_contract | Get the typed contract (produces, consumes, allowed handoffs) for a role. | role |
role_claim | Stamp a session with an agent role. | session_id, role |
role_handoff | Validate and produce a typed handoff between roles. Throws on illegal transitions or missing keys. | from, to, payload |