Skip to main content

Fusion AI

These entities hold the work of Fusion AI. Fields marked Platform are set for you when you save.


Agent​

A saved list of questions that Fusion answers in order, on demand or on a schedule. Service: AI Agent (aiagent/v1).

FieldTypeRequiredWhat it is
_idStringNoThe agent id
nameStringYesThe name
messagesArrayYesThe questions, in order, each {"content": "..."}
cronStringNoA schedule, see Cron
enabledBooleanNoFalse to stop the schedule, defaults to true
iconStringNoDefaults to person-circle text-success
user, _user, tid, runAsPlatformWho saved it, your tenant, and who scheduled runs act as
{
"_id": "morning-gas-check",
"name": "Morning gas check",
"cron": "30 7 * * 1-5 * Europe/London",
"messages": [
{"content": "Did the ICE Endex TTF settlement prices arrive yesterday?"},
{"content": "Summarise the change in the TTF front month over the last week"}
]
}

Assistant​

Service: AI Assistant (aiassistant/v1/{source}).

FieldTypeWhat it is
_idStringThe assistant id
nameStringThe name
descriptionStringWhat it is for. Fusion uses this to choose the assistant
systemPromptStringThe instructions given to Fusion when it uses the assistant
iconStringThe icon
modelStringThe Claude model, if not the default
effortStringThe effort level, if not the default
onlyUseCustomToolsBooleanTrue to use only your custom tools
{
"_id": "EPR",
"name": "EPR Assistant",
"icon": "recycle",
"description": "Answers questions about extended producer responsibility obligations",
"systemPrompt": "You help the Northgate compliance team with their EPR obligations..."
}

Playbook​

A playbook is a markdown document that describes a piece of work step by step. Service: AI Playbook (aiplaybook/v1/{source}), versioned.

To save a playbook, send just the markdown in source. The id comes from its front matter, and the other fields are worked out from it.

{"source": "---\nid: ttf-curve-review\nname: TTF curve review\n---\n..."}
FieldTypeWhat it is
_idStringFrom the front matter. Public ids start with #, private ids must not
sourceStringThe markdown
name, category, description, assistant, iconStringFrom the front matter
tagsString[]From the front matter
featuredBooleanShown as featured in the library
templateObjectThe input fields, the plan of steps and the outputs
stepCount, inputCount, gateCount, checkCount, transitionCountIntegerCounts of steps, inputs, approval gates, checks and transitions
toolsString[]The tools the steps can use
changesPlatformBooleanTrue if a step can change platform data
maxTokensIntegerThe most tokens a run may use
warningsString[]Warnings from the markdown, with line numbers
copiedFromObjectFor a copy of a library playbook: playbook and version

See the playbook syntax for the markdown.


Task​

A playbook with its values filled in for one piece of work, or scheduled to run regularly. Service: AI Task (aitask/v1).

FieldTypeRequiredWhat it is
_idStringNoThe task id
playbookStringYesThe playbook id
versionIntegerNoThe playbook version, defaults to the latest
valuesObjectNoThe values for the playbook's inputs, by input name
scheduleStringNoA schedule in the cron format with 6 or 7 parts
enabledBooleanNoFalse to pause a scheduled task
ownerStringNoDefaults to you
playbookName, fields, plan, outputsPlatformCopied from the playbook version
statusStringPlatformdraft, or scheduled for a task with a schedule
runAsObjectPlatformFor a scheduled task: the person each run acts as
lastRunPlatformSet once the task has run. A task that has run cannot be changed
{
"_id": "weekly-ttf-review",
"playbook": "ttf-curve-review",
"values": {"curve": "NGE.GAS.TTF:SETTLE"},
"schedule": "0 8 * * 1 * Europe/London"
}

Run​

The record of one playbook run, written by Fusion AI. Service: AI Run (airun/v1), read-only.

FieldTypeWhat it is
_idStringThe run id
playbookStringThe playbook run
userStringWho started it
statusStringrunning, awaiting_approval, done, failed, rejected, cancelled or stopped
modeStringinteractive or background
stepsArrayEach step's result, checks, approvals and error
outputsObjectThe outputs
tokens, maxTokensIntegerThe billed tokens used, and the most allowed
errorStringWhy a failed or stopped run ended
scheduledFromStringFor a scheduled run: the task it was copied from
_createdDateTimeWhen it started

Tool​

A function in one of your scripts that Fusion can call. Service: AI Tool (aitool/v1).

FieldTypeWhat it is
_idStringThe tool id
nameStringThe function name: letters, numbers, _ and -
scriptStringThe private script, with category aitool, that has the function
descriptionStringWhat the tool does. Fusion uses this to decide when to call it
propertiesArrayThe inputs, each with name, type, description and required
assistantStringLimits the tool to one assistant
iconStringThe icon
{
"_id": "nge-position",
"name": "nge_position",
"script": "northgate\\ai\\position-tools",
"description": "Returns the Northgate net position for a product and month",
"properties": [
{"name": "product", "type": "string", "description": "The product, for example TTF", "required": true},
{"name": "month", "type": "string", "description": "The delivery month as yyyy-MM", "required": true}
]
}

Conversation​

A conversation in the Fusion AI studio, saved as you chat. Service: AI Studio (aistudio/v1).

FieldTypeWhat it is
_idStringThe conversation id
nameStringThe name of the conversation
categoryStringA category
iconStringThe icon, chat-right-text
messagesArrayThe messages: role (user, assistant or log), content, assistant (which assistant answered), timestamp, usage (the tokens used) and feedback
user, _userStringWhose conversation it is
agentStringFor an agent run: the agent id, instead of a user
tidStringYour tenant
_created, _updatedDateTimeWhen it started and last changed

A conversation you save yourself must have a prompt (the first question) and text.


Usage​

Token usage, one record per conversation per day. Service: AI Usage (aiusage/v1), read-only.

FieldTypeWhat it is
conversationStringThe conversation id
userStringWho had the conversation
tidStringThe tenant id
dateStringThe day
timestampDateTimeWhen the record was last updated
input, outputLongInput and output tokens
cacheWrite, cacheReadLongTokens written to and read from the prompt cache
billedTokensLongThe billed tokens, see Billing

Feedback​

A rating of a Fusion AI answer. Service: Feedback (feedback/v1).

FieldTypeWhat it is
_idStringA UUID
conversationStringThe conversation id
assistantStringThe assistant that answered
questionStringThe question asked
contentStringThe answer rated
messageStringThe comment given with the rating
positiveBooleanTrue for a thumbs up, false for a thumbs down
user, _user, tidStringWho gave it, and your tenant
_createdDateTimeWhen it was given