📖 Reference

Reference

Complete tool index, signal types, data model, and integration reference for agents that learn. Shared across all lanes.

Signal Types

Every event has a signal_type that classifies what kind of knowledge it represents. Eight types cover the full range.

Type	What it represents	Value meaning	Lanes
competence	Skill or knowledge level in a domain	0 = novice, 1 = expert	Chat, Agent, Coding
preference	Communication or behavioral setting	0 = low/off, 1 = high/on	Chat, Agent, Coding, Devices
claim	A tracked proposition or belief	0 = false/rejected, 1 = true/accepted	Agent, Coding, Workflow, Multi-agent
entity	Named entity — person, org, tool, device	0 = inactive, 1 = active/relevant	All lanes
event	Incident, milestone, decision, anomaly	0 = minor, 1 = significant	Coding, Workflow, Devices
metric	Measurable quantity	Normalized 0–1 within expected range	Coding, Workflow, Devices
pattern	Recurring behavior or trend	0 = weak, 1 = strong pattern	Chat, Coding, Workflow, Devices
state	Current status or operational phase	Context-dependent (document in gloss)	Workflow, Multi-agent, Devices

Certainty & Learning

Every event carries a certainty score between 0 and 1. It represents how confident the model is in what it knows about this signal.

Certainty range	Meaning	Behavior
0.0 – 0.3	Low — new or weakly supported	Updates fast; open to revision
0.3 – 0.6	Medium — building confidence	Updates at moderate rate
0.6 – 0.8	High — well-supported	Resists noise; updates on strong signals
0.8 – 1.0	Very high — solidly established	Highly stable; only major contradictions move it

Start new observations at 0.3 and let learning refine over time. The system is agile where ignorant and stable where certain — learning rate adapts automatically to current certainty.

Validation types

Type	Use when
`direct`	A sensor or data source provides objective ground truth (compare predicted vs. actual to compute error)
`logic`	No external sensor — error magnitude is your judgment based on corroboration, contradiction, or specificity

Error magnitude guidance (for `apply_learning`)

Signal type	Magnitude	Direction
Independent corroboration	0.05 – 0.1	increase
Same-source repetition	0.02 – 0.05	increase
Specificity upgrade	0.1 – 0.15	increase
Temporal confirmation	~0.1	increase
Contradiction	0.2 – 0.4	decrease

Read Tools

Tool	Purpose	Key params
`get_pod_summary`	Condensed readable overview — top signals with glosses and freshness	`user_id`, `include_glosses`, `include_freshness`
`get_pod`	Full pod state including all events	`user_id`, `include_vectors`, `include_curiosity`
`get_events`	All events, optionally filtered by signal_type	`user_id`, `signal_type`, `validation_type`
`get_relevant_context`	Events most relevant to a task — scored by match + certainty	`user_id`, `query`, `max_results`
`query_pod`	Natural language question across all events	`user_id`, `query`
`get_low_certainty`	Events below a certainty threshold — best validation candidates	`user_id`, `threshold`
`get_curiosity_triggers`	Events that need validation, are stale, or under-observed	`user_id`, `max_triggers`, `include_probes`
`get_gap_analysis`	High-value, low-certainty signals ranked by investigation priority	`user_id`, `max_gaps`
`get_convergence_report`	New events, certainty changes, and stale signals since last report	`user_id`, `since_days`

Write Tools

Tool	Purpose	Key params
`create_pod`	Initialize a new pod for a user or entity	`user_id`, `name`, `pod_type`
`save_event`	Save or update any event	`user_id`, `key`, `signal_type`, `value`, `certainty`, `gloss`, `meta`
`record_observation`	Log a raw behavioral signal (correction, engagement, etc.)	`user_id`, `session_id`, `signal_type`, `signal_data`
`apply_learning`	Update an event's value and certainty based on an observation	`user_id`, `key`, `signal_type`, `error_direction`, `error_magnitude`
`update_session`	Increment session and interaction counters — call once at end	`user_id`, `interactions`
`delete_event`	Remove a specific event by key	`user_id`, `key`, `signal_type`
`delete_pod`	Permanently delete pod and all data — irreversible	`user_id`, `confirm: true`

Graph Tools

Tool	Purpose	Key params
`link_events`	Create a semantic relationship between two events	`user_id`, `source_key`, `target_key`, `relationship`, `note`
`get_event_cluster`	BFS traversal from an event through its links	`user_id`, `key`, `signal_type`, `max_depth`
`suggest_links`	Auto-suggest relationships based on keyword overlap	`user_id`, `min_overlap`
`detect_contradictions`	Surface conflicting high-certainty events	`user_id`, `certainty_threshold`
`suggest_synthesis`	Find high-certainty events that could be consolidated	`user_id`, `min_certainty`, `min_overlap`
`export_cluster_diagram`	Mermaid diagram of event relationships	`user_id`, `center_event`, `include_all`
`export_argument_outline`	Hierarchical markdown outline from linked events	`user_id`, `topic`, `include_evidence`
`get_argument_chain`	Trace logical dependencies for a claim or topic	`user_id`, `topic`, `max_depth`

Extended Context Tools

Tool	Purpose
`save_agent_context` / `get_agent_context`	Persist and retrieve agent task history, decisions, errors per agent_id
`save_repo_context` / `get_repo_context`	Codebase metadata across sessions, namespaced by repo
`save_service_context` / `get_service_context`	External service health and status data
`save_life_context` / `get_life_context`	User-shared life patterns (health, goals, routine, schedule)
`link_resource` / `get_resources` / `unlink_resource`	Attach and retrieve files, URLs, and projects linked to events
`add_source`	Register a data source with a credibility score for intelligence pipeline pods
`export_pod`	Export complete pod data as JSON (GDPR Article 20)

Relationship Types

Relationship	Meaning	Example
`BUILDS_ON`	A is grounded in B	`react` BUILDS_ON `javascript`
`ENABLES`	A makes B possible	`python` ENABLES `data-science`
`RELATED_TO`	Thematic connection	`hiking` RELATED_TO `photography`
`CONTRADICTS`	A conflicts with B	`minimalism` CONTRADICTS `feature-creep`
`PART_OF`	A is a subset of B	`flask` PART_OF `python-web`
`EVOLVES_FROM`	A grew out of B	`microservices` EVOLVES_FROM `monolith`
`SUPERSEDES`	A replaces B	`v2-api` SUPERSEDES `v1-api`

Event Schema

All events share a common schema regardless of signal type.

Event fields

{
  "user_id":          "string",          // pod identifier — required
  "key":             "string",          // signal key — required
  "signal_type":     "competence|preference|claim|entity|event|metric|pattern|state",
  "value":           0.0–1.0,           // signal strength — required
  "certainty":       0.0–1.0,           // model confidence — default 0.3
  "gloss":           "string",          // human-readable interpretation
  "validation_type": "direct|logic",    // how error signals are produced
  "meta":            {},                 // arbitrary structured metadata
  "source_id":       "string",          // source identifier (intelligence pods)
  "source_type":     "api|sensor|document|..."
}

MCP Config

Nervous Machine is a remote MCP server. Add it via your client's connector UI or config file.

claude_desktop_config.json / .cursor/mcp.json

{
  "mcpServers": {
    "nervous-machine": {
      "url": "https://context.nervousmachine.com/mcp",
      "transport": "sse"
    }
  }
}

ℹ️

Because Nervous Machine is a remote MCP server, most clients let you add it through their UI without editing config files directly. Check your client's "Add Integration" or "MCP Connectors" section.

curl Examples

Create a pod

curl -X POST https://context.nervousmachine.com/mcp \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"method":"tools/call","params":{"name":"create_pod","arguments":{"user_id":"alice","name":"Alice"}}}'

Save an event

curl -X POST https://context.nervousmachine.com/mcp \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"method":"tools/call","params":{"name":"save_event","arguments":{"user_id":"alice","key":"python","signal_type":"competence","value":0.8,"certainty":0.3,"gloss":"Strong Python developer"}}}'

Get pod summary

curl -X POST https://context.nervousmachine.com/mcp \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"method":"tools/call","params":{"name":"get_pod_summary","arguments":{"user_id":"alice","include_glosses":true}}}'

Apply learning

curl -X POST https://context.nervousmachine.com/mcp \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"method":"tools/call","params":{"name":"apply_learning","arguments":{"user_id":"alice","key":"python","signal_type":"competence","error_direction":"increase","error_magnitude":0.1}}}'

Privacy & Data Rights

Nervous Machine stores learned abstractions, not raw data. Your pod contains what the agent learned from interactions — not transcripts, not files, not personal information beyond what you explicitly save.

Right	Tool	Notes
Export your data	`export_pod`	Returns complete pod as JSON. GDPR Article 20 compliant.
Delete your data	`delete_pod`	Permanent, irreversible. Requires `confirm: true`. GDPR Article 17 compliant.
Inspect what's stored	`get_pod_summary`, `get_events`	Full visibility into all events and certainty scores at any time.
Correct stored knowledge	`save_event`, `apply_learning`, `delete_event`	Any event can be updated, corrected, or deleted in natural language.

✦

Vectors, not transcripts. Think of it like a colleague who remembers you're great at Python and prefer direct feedback. They don't have a transcript of your conversations — they just know you. That's what your pod stores.

← Previous Devices & Ecosystems ↑ Back to All Docs

Reference

Signal Types

Certainty & Learning

Validation types

Error magnitude guidance (for apply_learning)

Read Tools

Write Tools

Graph Tools

Extended Context Tools

Relationship Types

Event Schema

MCP Config

curl Examples

Privacy & Data Rights

Error magnitude guidance (for `apply_learning`)