Skip to main content
Glama

sage_remember

Stores durable facts, session observations, inferences, and tasks as consensus-validated memories, with replacement-based corrections to keep context accurate across agents.

Instructions

Store a memory in SAGE. When domain is omitted, app-v23 uses this agent's approved owned home domain (older nodes retain the legacy general default); an explicit domain is never remapped. For a correction, pass replaces_memory_id here instead of calling sage_forget first: SAGE stores and verifies the replacement before it challenges the old memory, so interruption can leave both records but can never leave neither. IMPORTANT: Use type='fact' (confidence 0.95) for durable knowledge that should persist long-term and be visible across all agents — infrastructure details (IPs, hostnames, SSH commands, URLs, ports), architecture decisions, verified configurations, credentials paths, and server specs. Use type='observation' for ephemeral session context. Facts survive confidence decay and cross provider boundaries; observations do not.

Tool call template

Replace each placeholder with a value matching the input schema. Only required parameters are included; add optional parameters as needed.

{
  "id": 1,
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "arguments": {
      "content": "<content>"
    },
    "name": "sage_remember"
  }
}

Input Schema

NameRequiredDescriptionDefault
tagsNoUser-defined labels for this memory (e.g. 'important', 'project-x')
typeNoMemory type. When omitted, a new memory defaults to observation and a correction inherits the original type. fact (0.95+): verified durable knowledge — IPs, hostnames, architecture decisions, configs, infrastructure. observation (0.80): session-level context — what happened, what was discussed. inference (0.60): hypotheses and conclusions. task: actionable items.
domainNoDomain tag. When omitted, a correction inherits its source domain and a new memory uses this app-v23 agent's owned home domain (legacy nodes use general). Explicit values are never silently remapped.
contentYesThe memory content to store: one self-contained statement a later reader can act on without this conversation (name the system or person; give the reason for a decision)
evidenceNoOptional source text this memory is based on (a quote, log line or document excerpt, up to 32 KiB). It stays on this node and never enters the chain; a node running the memory gate checks that it supports the memory as stated. Not for tasks.
confidenceNoConfidence score 0-1. When omitted, defaults to 0.90 for tasks (including inherited correction types), otherwise 0.80. Explicit values are preserved.
derived_fromNoAuthenticated app-v30 parent refs. Each parent must be committed or challenged, carry the exact content hash, and be readable by this caller.
replacement_reasonNoOptional audit reason recorded when the replaced memory is challenged.
replaces_memory_idNoOptional committed memory ID this content corrects. The replacement is committed first; only then is the old memory challenged.

Schema Changelog

Changes observed during successful MCP inspections.

  1. Changed2 schema fields changedv11.25.2
    • changedInput schema / properties / content / description
      Previous value: -"The memory content to store"New value: +"The memory content to store: one self-contained statement a later reader can act on without this conversation (name the system or person; give the reason for a decision)"
    • addedInput schema / properties / derived_from
      Added value: +{
      +  "description": "Authenticated app-v30 parent refs. Each parent must be committed or challenged, carry the exact content hash, and be readable by this caller.",
      +  "items": {
      +    "additionalProperties": false,
      +    "properties": {
      +      "content_hash": {
      +        "pattern": "^[0-9a-f]{64}$",
      +        "type": "string"
      +      },
      +      "memory_id": {
      +        "maxLength": 512,
      +        "minLength": 1,
      +        "type": "string"
      +      },
      +      "source_chain_id": {
      +        "maxLength": 64,
      +        "minLength": 1,
      +        "type": "string"
      +      }
      +    },
      +    "required": [
      +      "source_chain_id",
      +      "memory_id",
      +      "content_hash"
      +    ],
      +    "type": "object"
      +  },
      +  "maxItems": 16,
      +  "minItems": 1,
      +  "type": "array"
      +}
  2. Changed6 schema fields changedv11.23.24
    • removedInput schema / properties / confidence / default
      Removed value: -0.8
    • changedInput schema / properties / confidence / description
      Previous value: -"Confidence score 0-1"New value: +"Confidence score 0-1. When omitted, defaults to 0.90 for tasks (including inherited correction types), otherwise 0.80. Explicit values are preserved."
    • addedInput schema / properties / confidence / maximum
      Added value: +1
    • addedInput schema / properties / confidence / minimum
      Added value: +0
    • removedInput schema / properties / type / default
      Removed value: -"observation"
    • changedInput schema / properties / type / description
      Previous value: -"Memory type. A correction inherits the original type when omitted. fact (0.95+): verified durable knowledge — IPs, hostnames, architecture decisions, configs, infrastructure. observation (0.80): session-level context — what happened, what was discussed. inference (0.60): hypotheses and conclusions. task: actionable items."New value: +"Memory type. When omitted, a new memory defaults to observation and a correction inherits the original type. fact (0.95+): verified durable knowledge — IPs, hostnames, architecture decisions, configs, infrastructure. observation (0.80): session-level context — what happened, what was discussed. inference (0.60): hypotheses and conclusions. task: actionable items."
  3. Changed1 schema field changedv11.23.11
    • addedInput schema / properties / evidence
      Added value: +{
      +  "description": "Optional source text this memory is based on (a quote, log line or document excerpt, up to 32 KiB). It stays on this node and never enters the chain; a node running the memory gate checks that it supports the memory as stated. Not for tasks.",
      +  "type": "string"
      +}
  4. Changed2 schema fields changedv11.16.2
    • removedInput schema / properties / domain / default
      Removed value: -"general"
    • changedInput schema / properties / domain / description
      Previous value: -"Domain tag (e.g. general, security, code). A correction inherits the original domain when omitted."New value: +"Domain tag. When omitted, a correction inherits its source domain and a new memory uses this app-v23 agent's owned home domain (legacy nodes use general). Explicit values are never silently remapped."
  5. Addedv11.14.2
  6. Removedv11.13.8
  7. Addedv11.13.0
  8. Removedv11.9.2
  9. Addedv8.1.0
  10. Removedv8.0.0
  11. Addedv7.6.1
  12. Removedv7.5.9
  13. Changed1 schema field changedv6.6.9
    • addedInput schema / properties / type / description
      Added value: +"Memory type. fact (0.95+): verified durable knowledge — IPs, hostnames, architecture decisions, configs, infrastructure. observation (0.80): session-level context — what happened, what was discussed. inference (0.60): hypotheses and conclusions. task: actionable items."
  14. First observedv4.5.2

TDQS

A4.6/5.0
Behavior4/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

With no annotations, the description carries the full burden and mostly delivers: it discloses the domain defaulting rule (owned home domain vs legacy general, never remapped), the write-ordering guarantee for corrections (replacement committed and verified before the old memory is challenged, so interruption can leave both but never neither), and that facts survive confidence decay and cross provider boundaries. It does not cover permission/auth requirements, rate limits, or what a failed write returns, which keeps it below a 5.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Front-loaded with purpose, then domain behavior, then the correction path, then type semantics — a sensible decision order. It loses a point for redundancy: the fact-versus-observation guidance is stated in the description and again in the type property, and the list of infra examples is long enough to bloat the paragraph.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a 9-parameter, annotation-free mutation tool with no output schema, the description covers the decisions an agent must make: what to store, which type, what happens to domain and confidence defaults, and how to correct an existing memory safely. The only real omission is derived_from, a power-user parameter that the schema documents adequately for its narrow use.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 100%, so baseline is 3, but the description adds meaning the schema does not: facts are visible across all agents and cross provider boundaries while observations are not, and the correction flow explains why replaces_memory_id exists as a parameter rather than a separate call. It leaves derived_from entirely unexplained, which the schema only partially covers.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

Opens with a specific verb+resource ("Store a memory in SAGE") and immediately differentiates itself from the sibling sage_forget by framing the correction path. An agent can tell this is the write path for new memories versus recall/forget/link without opening any schema.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Explicitly states when to use type='fact' versus 'observation', with concrete examples of what counts as durable knowledge, and gives a hard routing rule: use replaces_memory_id here instead of calling sage_forget first. It names the alternative sibling and the condition that selects against it, which is the strongest form of usage guidance.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.