Conxt

Get Context

Load relevant memories from your Conxt graph for the current task.

Get Memories

List personal memories from your Conxt graph, with optional filters.

Search Decisions

Search your decisions and coding rules for a specific topic.

Get Pinned

Get your pinned memories — the most important permanent context.

Add Memory

Save a new memory to your Conxt graph. Pass `title` as a short imperative summary under 120 characters and `content` as the full explanation. Pass `tags` as a list of short strings. Do NOT write a title, tags, or a confidence value inside `content` — each has its own parameter, and burying them in the text corrupts the similarity embedding used for deduplication. Confidence is not settable over MCP; all writes are stored at 0.9. IMPORTANT FOR AGENTS: always pass your own agent_id (e.g. "cursor

List My Teams

List my teams / team workspaces and their team IDs. Returns every team you belong to (owned or member) with its team_id. Call this FIRST to get a team_id before using any team-scoped tool (get_team_memories, add_team_memory, propose_memory, set_memory_status). Use when you need to find a team, look up a team id, or see which team workspaces you have access to.

Get Team Memories

Read memories belonging to a specific team workspace.

Add Team Memory

Save a memory to a team workspace (visible to all team members). Pass `title` as a short imperative summary under 120 characters and `content` as the full explanation. Pass `tags` as a list of short strings. Do NOT write a title, tags, or a confidence value inside `content` — each has its own parameter, and burying them in the text corrupts the similarity embedding used for deduplication. Confidence is not settable over MCP; all writes are stored at 0.9. Team writes require a real credential —

Propose Memory

Propose a memory to a team vault for human review. Creates a PENDING candidate. It is NOT loaded into any agent's decision context and does not govern anything until a human approves it with set_memory_status. This is the write verb an agent should use for a team/agent vault: unlike add_team_memory (which can reinforce an existing record via dedup), a proposal always lands as a fresh review candidate, and it never touches the approved vault it was read from. Returns a proposal_id so the caller

Set Memory Status

Approve, pause, or re-open a TEAM memory. approved — live; loaded into agent decision context and governs. paused — on hold; reversible; NOT loaded. Nothing is deleted. pending — send back to the review queue. Governing a team vault is a HUMAN action. This requires an interactive, signed-in session (the dashboard) — an agent authenticating with a cnxt_ API key can propose (propose_memory) but cannot approve, pause, or re-open, no matter whose credential it holds. That's what makes

Update Memory

Revise an existing memory. `content` replaces the memory text. `title` and `tags` are optional and are overwritten only when supplied — omit them to leave existing values alone. Note this repairs the stored record, not its embedding: the vector is generated at insert and is not recomputed here. A memory whose text changes materially is better archived and re-added.

Delete Memory

Retract a memory (soft delete).

Resolve Context

Server-side verified retrieval loop -- runs DISCOVER, VERIFY, and ITERATE internally and returns ONE checked answer, instead of raw results the caller has to judge for itself. This is the difference between a tool that hands back candidates and a tool that hands back an answer it can stand behind. The loop, concretely: 1. DISCOVER -- semantic search over memory for the query. 2. VERIFY -- checked against REAL signals already in the schema, not model judgment: - empty result?

Search Memory

Semantic search over your memory, tuned for agent recall.

Recall As Of

Recall memories about a topic, sorted PURELY by recency rather than hybrid_score's blended similarity+recency+reinforcement ranking. This is the direct fix for a staleness gap get_context doesn't cover: hybrid_score can let an older, heavily-reinforced memory outrank a newer, correct one, if the older one was reinforced many times before it was superseded. recall_as_of ignores reinforcement and confidence entirely -- it answers "what's the MOST RECENT thing you know about X", or "what did you k

List Agent Memories

List memories written by an agent (or in a specific session).

Get Full Context

Consolidate EVERYTHING known about a topic into one grouped answer. Use this instead of resolve_context or search_memory when the user asks to "pull together", "summarize everything about", "what do you know about", or otherwise wants a COMPLETE picture rather than a single fact. resolve_context returns one verified answer; this returns the full relevant set, grouped by memory type, in a single call — so an agent doesn't have to chain multiple searches to assemble a summary. Args: topic: