Authentication for the REST API using Clerk.
Turns an Obsidian vault into a memory store, with markdown files as memory nodes and wikilinks as causal edges.
Provides embeddings for memory storage and recall via OpenAI API.
Persistent storage backend using PostgreSQL with pgvector for vector search.
Graph-native storage via FalkorDB, a Redis-based graph database.
The intelligence layer for AI memory.
Papez doesn't just remember what happened; it remembers why. A scoring engine + causal graph + lifecycle manager for AI agent memory. Speaks MCP natively.
System | Score | Protocol |
Papez | 85.55 ± 0.37 | Frozen: gpt-4o-mini answerer + judge, temp 0, n=1,540, cats 1–4, 10 runs (July 2026) |
Zep | 75.14 | Comparable published setup |
Mem0 | 66.9 | Comparable published setup (Mem0 paper) |
Self-reported vendor figures above ~90 use different answerers/judges and are not comparable — the oracle retrieval ceiling under this frozen protocol is 94.9. Reproduce it yourself: Astrix-Labs/locomo-harness · full methodology · per-run results.
Hosted product: papez.ai — your personal memory for AI, carried across ChatGPT, Claude, and every MCP app · Pricing · Developer docs · Benchmark methodology (85.55 on LoCoMo, certified over 10 runs, receipts published)
Related MCP server: tentra
Papez is a scoring engine, causal graph, and lifecycle manager for AI memory. Memories are scored by a multiplicative formula (relevance × connectivity × reactivation), connected in a causal graph, and actively forgotten when they become irrelevant.
This package (papez) is the core library: an in-memory causal graph engine with optional JSON persistence, plus a stdio MCP server. It has no database dependency and no REST API. A hosted product built on top of this library — with Postgres, additional storage backends, and a REST/HTTP MCP API — is available separately at api.papez.ai; it is not part of this package.
Flat memory doesn't scale. Dumping everything into a vector store gives you recall with zero understanding. The 500th memory buries the 5 that matter.
No forgetting = no intelligence. Real memory systems forget. Without active pruning, your AI drowns in stale context.
No causal reasoning. Vector similarity can't answer "why did I choose X?" — you need a graph.
Your AI remembers everything but understands nothing. Papez fixes that.
Requires Python 3.11 or newer.
Install the package. The base install has zero database dependencies — state lives in memory and is optionally persisted to a JSON file.
pip install papezOptional extras:
pip install 'papez[openai]' # OpenAI embeddings
pip install 'papez[local]' # Local embeddings (sentence-transformers, no API key)
pip install 'papez[anthropic]' # LLM-based causal inference (consolidation, contradiction detection)Run the stdio MCP server directly:
python3 -m papezgit clone https://github.com/Astrix-Labs/papez.git
cd papez
pip install -e '.[dev]'
pytest tests/claude mcp add papez -- python -m papezAdd to your claude_desktop_config.json:
{
"mcpServers": {
"papez": {
"command": "python",
"args": ["-m", "papez"]
}
}
}The stdio server is a single local process. Under load — or during a restart or redeploy of a hosted transport in front of it — a tool call can transiently fail or the connection can briefly go unresponsive. Memory writes and reads are not worth crashing an agent turn over, so clients should degrade gracefully rather than treat a memory call as fatal:
The server degrades gracefully too: a tool exception (or a missing
required argument) is returned as a structured
{"error": "...", "retryable": bool} payload instead of a protocol-level
MCP failure, so a memory hiccup never crashes the transport. The
retryable flag encodes the guidance below — true only for read tools.
Retry idempotent reads (memory_recall, memory_search, memory_traverse,
memory_explain, memory_stats) with a short bounded backoff (e.g. 2–3
attempts). These have no side effects worth worrying about beyond reactivation
bookkeeping.
Do not blindly retry memory_store / memory_amend on an ambiguous timeout
— a silent success followed by a retry creates a duplicate node. Prefer to
continue the turn and reconcile on the next memory_recall, or pass a stable
source_session so duplicates are easy to spot.
Treat memory as best-effort context, not a hard dependency. If a call fails, proceed with whatever context you already have and try again next turn rather than aborting. The graph is durable; a missed write is recoverable, a crashed agent turn is not.
Tool | Description |
| Store a new memory. Use |
| Record a correction: creates a new memory that supersedes an existing one. The old memory is kept (decayed in recall), not deleted. |
| Recall memories by natural language query (vector + keyword + graph spreading activation). Supports |
| Filtered vector search by status, category, date ( |
| Walk the causal graph from a node. Returns reachable nodes and the edges of the induced subgraph ( |
| Explain a memory's score. Includes a |
| Get memory system statistics |
| Pin a memory so it's never forgotten |
| Unpin a previously pinned memory |
| Permanently delete a memory |
| List core memories, optionally filtered by category |
| Set user preferences for core memory categories |
| Promote a private memory to org visibility |
memory_store's related argument lets the writer set edge semantics instead of
guessing. Each entry is {"id": "<node-id>", "type": "<edge-type>"}, directed
new_node --type--> target (so supersedes means the new node supersedes the
target). Invalid types are rejected before the node is created — explicit
writes never half-succeed. related_to still exists but always creates
caused_by; prefer related.
To correct a fact, use memory_amend(node_id, content, reason=...): it stores the
new version, links it SUPERSEDES → old, and keeps the old memory for audit.
Recall automatically deprioritizes superseded hits and tags them with
superseded_by.
When you memory_store something that lexically disagrees with an auto-link
candidate (a changed number, a negation), the result may include
possible_conflicts — heuristic hints, not verified contradictions, and never
materialized as edges. Use them to decide whether to memory_amend.
memory_recall(query, verbosity="concise") skips the causal-chain enrichment and
returns only id / summary / status / score / activation / is_core (plus
superseded_by when set) per hit — much cheaper on tokens for high-frequency
lookups. verbosity="full" (the default) is unchanged. Reactivation writes still
occur in both modes (they are governed by read_only, not verbosity).
See docs/scoring.md for what activation / decay_score
actually mean — in short, it is a retention weight that rises when a memory is
recalled, not a countdown to deletion.
Every memory is scored by three forces multiplied together:
decay_score = relevance × connectivity × reactivationRelevance decays over time. Old memories fade unless reinforced.
Connectivity rewards memories with many causal links. Hub memories survive.
Reactivation boosts memories that keep getting recalled. Frequency matters.
Because the formula is multiplicative, a memory must score on all three axes to survive. A highly connected but never-accessed memory still decays. A frequently recalled but causally orphaned memory still fades.
decay_score (aliased activation on every hit) is a retention weight, not a deletion countdown — recalling a memory raises it, and a low score just means "resting," not "doomed." Deletion requires a low score and orphaned and unpinned and non-core and non-org and idle (not stored, recalled or reactivated for 30 days, GENESYS_FORGETTING_MIN_IDLE_DAYS), all at once. The stdio server runs the rescore-transition-prune pass every 10 minutes (GENESYS_MAINTENANCE_INTERVAL_S; 0 disables it). See docs/scoring.md for the full model and worked numbers.
STORE → ACTIVE → DORMANT → FADING → PRUNED
↑ │
└── reactivation ────┘
(only if score=0, orphan, not pinned)Memories can also be promoted to core status — structurally important memories that are auto-pinned and never pruned.
See the certified LoCoMo results at the top of this README: 85.55 ± 0.37 over 10 runs under a frozen protocol (gpt-4o-mini answerer and judge, temperature 0, n=1,540, categories 1–4). Category 5 — adversarial questions with disputed ground truth — is excluded, matching the comparable published setups.
Every run is reproducible: the harness is at Astrix-Labs/locomo-harness, with full methodology and per-run results published. Reproduction scripts for the in-repo scenarios are in benchmarks/.
This package ships one storage backend: an in-memory causal graph (storage/memory.py) with optional JSON persistence via GENESYS_PERSIST_PATH. No database is required.
Additional backends — Postgres/pgvector, FalkorDB, MongoDB, and an Obsidian vault adapter — along with a REST API, OAuth, and multi-user auth, are part of the hosted product at api.papez.ai and are not included in this repo.
Want a different storage backend for the open-source library? Implement the provider protocols in storage/base.py and bring your own.
Copy .env.example to .env and set:
Variable | Required | Description |
| Unless | Embeddings |
| No | Enables LLM-based causal inference (consolidation, contradiction detection). Off by default — without it, causal edges only come from edges the caller explicitly declares in |
| No |
|
| No | JSON file path to persist state across restarts (in-memory otherwise) |
| No | Default user ID for single-tenant mode |
Auto-linking connects a newly stored memory to semantically similar existing memories. If it is too permissive you get a "hairball" — everything ends up ~2 hops from everything, which destroys traversal scoping. Three knobs bound it:
Variable | Default | Description |
| embedder-recommended | Cosine floor to create an auto-link. Explicit value wins over the embedder default. |
|
| Max auto-links a single |
|
| Max |
The floor is embedder-aware: an auto-link is permanent graph structure, so its
floor sits above the transient recall floor. When GENESYS_AUTOLINK_MIN_SIMILARITY
is unset, the effective floor is the embedder's recommendation — 0.6 for OpenAI
(text-embedding-3-small, whose genuine matches cluster ~0.5+) and 0.45 for
local sentence-transformers (whose genuine matches cluster ~0.2–0.4 but whose
noise pairs have been observed at ~0.44, so only near-duplicate content
auto-links locally). Any unknown embedder falls back to 0.45. Auto-linking also
de-dupes: if a pair is already connected by any edge (e.g. a user_explicit
caused_by), no parallel auto_link related_to is created.
The possible_conflicts hint on memory_store scans with its own, lower
floor (GENESYS_CONFLICT_MIN_SIMILARITY, defaulting to the recall floor) over
a wider window (GENESYS_CONFLICT_SCAN_K, default 8) — so tightening the
auto-link floor never shrinks conflict detection.
The same embedder-aware pattern governs recall filtering:
Variable | Default | Description |
| embedder-recommended (OpenAI 0.5 / other 0.2) | Cosine floor below which pure vector hits are dropped from |
| embedder-recommended (OpenAI 0.45 / other 0.2) | Floor for injecting auto-promoted core memories into recall results. Pinned memories are always injected. |
The three-force scoring formula and its lifecycle thresholds are all
env-configurable (see engine/config.py and
docs/scoring.md): GENESYS_ACTR_DECAY,
GENESYS_RELEVANCE_VECTOR_WEIGHT, GENESYS_RELEVANCE_KEYWORD_WEIGHT,
GENESYS_MIN_CONNECTIVITY, GENESYS_FORGETTING_THRESHOLD, the GENESYS_DORMANCY_*
transition thresholds, and the GENESYS_CORE_* promotion weights.
See .env.example for all options.
Papez is built by Rishi Meka at Astrix Labs. It came out of frustration with re-explaining project context to Claude every session. The goal is the intelligence layer between your LLM and your memory — fully open source.
See CONTRIBUTING.md.
Note: Papez releases prior to v0.3.6 were documented as Apache 2.0 in error. The LICENSE file has always contained the AGPLv3 text. From v0.3.6 onward, all documentation correctly references AGPL-3.0-or-later with a Contributor License Agreement.
Permanently delete a memory node and all its edges.
| Name | Required | Description | Default |
|---|---|---|---|
| node_id | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations provided, the description carries the burden. It reveals permanence and that edges are removed, but does not mention authentication or rate limits. For a simple delete operation, this is adequate.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single, focused sentence that directly states the action and scope. There is no unnecessary detail or repetition.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a destructive operation, the description conveys the essential information: what is deleted (node and edges) and that it is permanent. No output schema is needed for such a simple action, and the description is complete enough for an agent to invoke it correctly.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 0% and the description does not elaborate on node_id. Although the parameter name is self-explanatory, the description could have added more context (e.g., format or example) to compensate for the lack of schema documentation.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the action (permanently delete) and the resource (memory node and all its edges). It is specific and distinguishes itself from sibling tools like memory_store or memory_amend.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description explains what the tool does but does not explicitly contrast with alternatives (e.g., when to delete vs. amend). However, the action is unambiguous and the sibling context implies usage.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
List all core memories, optionally filtered by category.
| Name | Required | Description | Default |
|---|---|---|---|
| category | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description carries the full burden. 'List' unambiguously indicates a read-only operation with no side effects, providing full behavioral transparency.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single concise sentence with no redundant information or extra clutter.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
The description provides sufficient context for an agent to decide to use it: it lists all core memories with an optional category filter, which is complete for a listing operation.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The 'category' parameter is described as an optional filter, giving it clear meaning beyond the bare schema type. It does not enumerate possible values, but the purpose is well understood.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the action (list) and the resource (core memories) with an optional filter, distinguishing it from other memory operations such as search or recall.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
No explicit comparison is made with sibling tools like memory_search or memory_recall, so an agent may not know when to prefer listing over searching.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Record a correction: creates a new memory that supersedes an existing one. The old memory is kept (decayed in recall results), not deleted.
| Name | Required | Description | Default |
|---|---|---|---|
| reason | No | ||
| content | Yes | ||
| node_id | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The description discloses that the old memory is kept but decayed in recall results, which is a behavioral detail beyond the basic schema. It does not mention error cases or effects on nodes, but the core behavior is transparent.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is one concise sentence that packs the purpose and key behavior without unnecessary words.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a simple tool, the description covers the primary purpose and effect, and the parameter names plus context allow an agent to infer usage. Missing explicit parameter meanings are a minor gap.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The schema has no parameter descriptions, but the description implies node_id identifies the existing memory. The fields 'reason' and 'content' are self-explanatory in context, but the tool description does not explicitly explain each parameter's role.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool records a correction and creates a new memory that supersedes an existing one, distinguishing it from the sibling memory_store tool which likely creates new memories without superseding.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
It implies usage for correcting existing memories by saying 'supersedes an existing one', but does not explicitly contrast with memory_store or other siblings. However, the intent is fairly clear.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Explain a memory's score breakdown.
| Name | Required | Description | Default |
|---|---|---|---|
| node_id | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations (e.g., readOnlyHint, destructiveHint) are provided, and the description does not mention side effects, permissions, or return behavior. It is unclear whether this tool only reads data or modifies anything.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single, concise sentence with no redundant words. It directly states the tool's purpose without unnecessary detail.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a simple tool this description is minimally adequate, but it lacks context about the output format or any additional behavior. The missing parameter details and return expectations reduce completeness.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The single parameter node_id has no schema description and is not elaborated in the tool description. This leaves ambiguity about what node_id refers to and how it should be supplied.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the action (explain) and the resource (a memory's score breakdown), distinguishing it from sibling tools like memory_store or memory_search. It is specific enough to understand the purpose.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
No explicit guidance is given on when to use this tool versus alternatives. The description implies use when wanting to understand a memory's score, but does not state conditions or prerequisites.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recall memories using hybrid search (vector + keyword + graph spreading activation).
| Name | Required | Description | Default |
|---|---|---|---|
| k | No | ||
| query | Yes | ||
| verbosity | No | concise = id/summary/status/score/activation/is_core only, no causal chains. | full |
| max_results | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description carries the full burden of explaining behavior, but it only mentions the hybrid search approach. It does not disclose return format, side effects, or how results are sorted/ranked.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single concise sentence that conveys core functionality without unnecessary words or repetition.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the complexity of the tool (hybrid search with multiple parameters) and the existence of closely related sibling tools, the description lacks detail about expected output, use cases, or how it differs from alternatives. It feels incomplete for an agent to choose confidently.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 25% (only 'verbosity' has a description). The description does not clarify the meaning or effect of 'query', 'k', or 'max_results' beyond what the schema implies.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool recalls memories using hybrid search, specifying the resource (memories) and the method (vector + keyword + graph spreading activation). It is distinguishable from siblings like memory_search, though not explicitly contrasted.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description provides no guidance on when to use this tool versus alternatives such as memory_search or memory_traverse. No criteria or context for selection is given.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Filtered vector search by status, category, date, or entity. Pass an EMPTY query to enumerate by recency instead (no vector search, no embedder needed): with since/active_since this answers 'what's new/changed since ' without knowing what to query for.
| Name | Required | Description | Default |
|---|---|---|---|
| k | No | ||
| query | Yes | Search text. Empty string switches to enumeration mode: nodes listed by last_reactivated_at descending, honoring the same filters. | |
| filters | No | Supported keys: status (list), category, entity, since (ISO date — created_at >= since; tz-naive treated as UTC), active_since (ISO date — last_reactivated_at >= active_since; tz-naive treated as UTC). With a non-empty query, results are vector-seeded and k-limited; use an empty query for enumeration. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations are provided, and the description does not mention whether the operation is read-only, whether it has side effects, or any auth or rate-limit considerations. Since it is a search tool, it likely does not modify data, but the description does not confirm this, leaving behavior partly unclear.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is concise and mixes the core purpose with the special enumeration mode efficiently. The punctuation and parenthetical asides are slightly complex, but the overall structure is clear and not verbose.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
The description covers the main usage modes and filter details, but it does not mention what the return values look like (no output schema) or how the `k` parameter affects the result set. For a search tool, this leaves some context about expected output incomplete.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The `filters` parameter receives a detailed explanation covering its keys and semantics, and the `query` parameter's empty-string behavior is well described. However, the `k` parameter is only given a type and default without any explanation of its meaning (e.g., maximum number of results), so parameter semantics are only partially covered.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the core function as a filtered vector search by status, category, date, or entity, and explicitly introduces the enumeration mode with an empty query. This makes the purpose immediately understandable and distinguishable from typical search tools.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description provides practical guidance on when to use the empty-query enumeration mode, especially for answering 'what's new/changed since <ts>' without needing a query. It does not explicitly compare with sibling tools, but the use cases are well enough defined for the primary scenarios.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Get graph statistics.
| Name | Required | Description | Default |
|---|---|---|---|
No parameters | |||
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
There are no annotations provided, so the description carries the full burden. It implies a read-only operation via 'Get', but does not explicitly state that it is non-destructive, what data it returns, or any side effects or permissions. The behavior is only minimally transparent.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single, concise sentence that directly states the purpose. It adheres to the principle of brevity and clarity, with no extraneous words or structure.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the lack of an output schema, the description is incomplete in explaining what the agent should expect from the tool. It does not say what kind of statistics are returned (e.g., counts, sizes, metadata) or how they might be used. An agent would need additional context to correctly interpret the result.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The tool has zero parameters, so the schema coverage is 100%. Per the baseline, a score of 3 is given when the schema fully documents all parameters. The description does not add any additional meaning about parameters, but none are needed.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description states a clear action 'Get' and an object 'graph statistics', but the object is vague. It does not specify what kind of statistics, which graph, or how they are presented. It is better than a tautology but lacks specificity to fully distinguish from similar tools like memory_explain or memory_recall.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description provides no guidance on when to use this tool versus its siblings. It does not mention any conditions, prerequisites, or alternatives. An agent would have to infer from the name alone that it is for retrieving statistics, with no clear differentiation from other memory tools.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Store a new memory in the causal memory graph. Use related for writer-specified typed edges (each {id, type}); related_to is legacy and always creates caused_by edges. May return possible_conflicts — heuristic hints, not verified contradictions.
| Name | Required | Description | Default |
|---|---|---|---|
| org_id | No | Required when visibility is 'org'. Must be an org the caller belongs to. | |
| content | Yes | ||
| related | No | Typed explicit edges. Direction: new_node --type--> target. | |
| category | No | Free-form classification (suggested: professional, educational, family, location). | |
| related_to | No | Legacy: ids of nodes to link via caused_by. Prefer `related`. | |
| visibility | No | private | |
| source_session | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations provided, the description carries the full transparency burden. It discloses the primary side effect (storing a new memory), the edge-creation behavior, legacy behavior of 'related_to', and notes that 'possible_conflicts' may be returned as heuristic hints. It does not mention authentication or permission side effects, but the core behaviors are transparent.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is concise, using only two sentences to cover purpose, edge semantics, legacy behavior, and return hints. There is no redundancy or unnecessary detail.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
The description covers the essential purpose, key parameter distinctions, legacy behavior, and return hints. It omits some details about fields like 'content' and 'visibility', but the overall context is sufficient for a typical agent to correctly invoke the tool.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The description adds value by explaining 'related' edges, the legacy nature of 'related_to', and the direction semantics. However, schema coverage is only 57%, and the description does not compensate for undocumented parameters like 'content', 'visibility', or 'source_session'. It partially clarifies parameters but not comprehensively.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool's primary action: storing a new memory in the causal memory graph. It distinguishes itself from sibling tools like memory_amend (existing memories), memory_recall, and memory_search by emphasizing 'new memory'.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description gives clear context for using the tool, including guidance on preferring the 'related' parameter over the legacy 'related_to' and clarifying that 'related_to' always creates caused_by edges. It does not explicitly say 'use this instead of memory_amend for existing memories,' but the context is clear enough.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Traverse the memory graph from a starting node. Returns reachable nodes AND the edges of the induced subgraph among them (source/target/type/weight/created_by) — a superset of the BFS tree, so paths can be reconstructed.
| Name | Required | Description | Default |
|---|---|---|---|
| depth | No | ||
| node_id | Yes | ||
| edge_types | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The description does add behavioral context: it states the return contains both reachable nodes and the edges of the induced subgraph, and explains it is a superset of the BFS tree. However, with no annotations present, it doesn't disclose whether traversal is read-only, whether there are cycle risks, or any side effects, so the full burden is not met.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description uses two sentences with no wasted words. It front-loads the main action, then immediately explains the output's structure and why the superset property matters. Very effective.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a tool with three parameters and no output schema or annotations, the description is incomplete. It leaves depth and edge_types undefined, and it doesn't explain what happens with empty reachability or missing node_id. This is not enough for a confident call.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The schema coverage is 0%, so the description must compensate for node_id, depth, and edge_types. It only mentions a starting node, which maps to node_id; it doesn't explain what depth controls or how edge_types filters traversal. This is a clear gap.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
This sentence is a precise action: the verb 'Traverse' with resource 'memory graph' and the starting node. It also clarifies the output is a superset of the BFS tree, which sets it apart from sibling tools like memory_recall or memory_search.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description does not say when to use this tool instead of alternatives, and it never names a sibling or a use case that would select this over memory_recall or memory_search. There is only an implicit purpose, no explicit when/when-not.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Pin a memory to core status.
| Name | Required | Description | Default |
|---|---|---|---|
| node_id | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Since no annotations accompany the tool, the description carries the full responsibility of explaining behavior. It only says 'Pin a memory to core status' and does not specify what 'core status' implies, whether the operation is reversible, what side effects might occur, or any permission-related constraints. The description effectively relies on the tool name itself.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is highly concise, with a single sentence and no filler. However, this conciseness is achieved by omitting nearly all the information an agent would need; it earns its place as a short purpose statement but does not go beyond that, making the brevity a trade-off against completeness.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
With no annotations, no output schema, and a bare schema with a single undocumented parameter, the description leaves central concepts undefined. An agent cannot infer what 'core status' means, how the memory is located via 'node_id', or what the tool actually does beyond the name. The description is not complete enough to support correct invocation.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The input schema has only one parameter, 'node_id', but its description coverage is 0%. The description does nothing to explain what 'node_id' represents, how to obtain it, or how it relates to the 'memory' being pinned. This is a critical gap because neither the schema nor the description supplies this required meaning.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description names a clear verb ('Pin'), a specific resource ('a memory'), and the intended outcome ('core status'), and it also implicitly distinguishes the tool from its sibling 'unpin_memory' by describing the opposite action. An agent can confidently connect the tool to its core function without consulting other tools.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description provides no guidance on when to use 'pin_memory' versus related tools such as 'memory_store', 'memory_amend', or 'unpin_memory'. It does not mention prerequisites, use cases, or exclusion conditions, so the agent receives no direction about when this tool is the right choice.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Promote a private memory to org visibility. Caller must own the node and belong to the target org.
| Name | Required | Description | Default |
|---|---|---|---|
| action | No | keep_private | |
| org_id | Yes | ||
| dry_run | No | ||
| node_id | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The main behavior is disclosed (changing a memory from private to org visibility) and the required permissions are stated. However, key behavioral controls in the schema—especially the 'action' enum values and 'dry_run' flag—are unexplained, and no annotations exist to fill that gap.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is two brief sentences with no redundant words or filler. It front-loads the core purpose and then states the key precondition, making it easy to parse quickly.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the lack of annotations and output schema, the description is not complete enough for reliable invocation. It omits the semantics of the 'action' enum, the behavior of 'dry_run', expected outcomes, and any edge cases or error conditions.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
With zero schema descriptions, the description must compensate. It implicitly covers node_id and org_id ('own the node', 'target org'), but it does not explain the meaning or effects of 'action' or 'dry_run'. Coverage is partial at best.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the primary action: promoting a private memory to org visibility. It also names the resource ('private memory') and the target state ('org visibility'), and it is distinct from sibling tools like pin_memory or memory_store.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description provides a precondition ('Caller must own the node and belong to the target org') but does not say when to use this tool versus alternatives like pin_memory or memory_store. No explicit usage guidance or comparison to sibling tools is given.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Configure core memory category preferences.
| Name | Required | Description | Default |
|---|---|---|---|
| auto | No | ||
| approval | No | ||
| excluded | No |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations and a generic 'configure' verb, the description does not disclose side effects, persistence, permissions, or whether changes are reversible. It is unclear if this tool modifies global settings or per-category rules.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single concise sentence with no redundancy. However, it sacrifices clarity for brevity, leaving out essential details, so it does not fully earn its place.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the lack of parameter explanations and usage context, the description is incomplete for an agent to safely and effectively invoke the tool. It does not cover return values, errors, or interactions with sibling tools.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The schema provides zero descriptions for parameters (auto, approval, excluded), and the description does not explain their meaning or expected values. The agent cannot determine what these array parameters control or how to populate them.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description states a verb ('Configure') and a resource ('core memory category preferences'), but it is vague about what 'preferences' entails. It does not clarify whether it sets auto-approval, exclusions, or other specific behaviors, making it only partially clear.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
No guidance is provided on when to use this tool versus siblings like memory_store or memory_amend. The description does not indicate scenarios where configuring preferences is appropriate, leaving the agent to infer usage.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Unpin a memory and re-evaluate core eligibility.
| Name | Required | Description | Default |
|---|---|---|---|
| node_id | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations or output schema. The description mentions side effects vaguely ('re-evaluate core eligibility') but does not explain what happens to the memory or return value.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Extremely concise and direct; no filler or redundant information.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
No output schema or return details, and the side effects of unpinning are under-specified, leaving the agent unsure about the outcome.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The only parameter node_id has no description in the schema, and the tool description does not clarify its format or role beyond the obvious identifier meaning.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
Clear verb 'unpin' and object 'memory', with an explicit consequence (re-evaluate core eligibility). Distinct from sibling tools like pin_memory.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
No guidance on when to use this tool versus alternatives such as pin_memory or delete_memory, nor any prerequisites.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
v0.1.0delete_memorylist_core_memoriesmemory_amendmemory_explainmemory_recallmemory_searchmemory_statsmemory_storememory_traversepin_memorypromote_to_orgset_core_preferencesunpin_memoryScored across 13 tools
Each tool targets a distinct operation: store, amend, recall, filtered search, graph traversal, explanation, pinning, unpinning, deletion, stats, preferences, and visibility promotion. memory_recall and memory_search are clearly differentiated as hybrid retrieval versus structured filtering/enumeration.
The memory_* prefix is used consistently for several core operations, but other tools switch to verb_memory forms (pin_memory, delete_memory), noun-like names (memory_stats), or unrelated forms (list_core_memories, set_core_preferences, promote_to_org). The naming is readable but not a uniform verb_noun pattern.
Thirteen tools is well within the ideal range for a memory graph server. Each tool covers a meaningful lifecycle or administrative function without unnecessary redundancy.
The tool set covers the full memory lifecycle: create, read via multiple retrieval modes, amend/supersede, pin/unpin, delete, and administrative operations like stats and preferences. Graph traversal and explanation tools add strong domain coverage with no obvious dead ends.
Graph-native persistent memory for AI agents — 33 MCP tools, zero-LLM writes.
Graph memory for AI agents: entities, cause-effect links, cross-session recall, time travel.
Persistent memory and knowledge graphs for AI agents. Hybrid search, context checkpoints, and more.
Persistent, inspectable memory for AI agents with lineage, correction, and a hosted MCP endpoint.
