AI Agents & Automations
Collate 2.0 replaces the fixed agent-app model with a general AI Automations system, renames the Skills API, folds Knowledge Center into the platform’s Context Center, and closes two long-standing access-control gaps on AI Dashboards and AI charts.The Documentation, Tier, and Quality agent apps are deleted
Breaking. Affects every deployment that installed the Documentation, Tier, or Quality agent.
Their
agentType enum values (CollateAI, CollateAITierAgent, CollateAIQualityAgent) are removed
from the schema — only Metadata remains — and each app’s bot is deleted along with it.
CollateAIApplicationBot and the Tier and Quality agent bots no longer exist after the upgrade, so
any token issued to one of them stops authorizing immediately.
The migration converts each installed app into an AI Automation carrying the original Elasticsearch
filter, named with a (migrated) suffix, but it arrives disabled and unscheduled — your
Documentation, Tier, and Quality automation stops running at the upgrade and stays stopped until
someone turns it back on. Fresh 2.0 installs never had the agent apps, so none of this applies to
them; the templates are simply available in the catalog from day one.
Automation that filters GET /v1/services/ingestionPipelines by the old agentType values must move
to the new values metadata and application.
AI Automations: a materially larger replacement surface
Additive, but the shape is different enough that anything built against the old agent-app APIs needs to change, not just repoint. AI Automations are a first-class entity at/v1/ai/automations with a considerably larger surface
than the apps they replace:
Run records are mapped onto the same shape the app run history used, so dashboards built on agent run
records keep working, but they must read from the automations endpoints, not
GET /v1/collate/apps/name/{name}/agentRuns, which is removed.Abilities renamed to Skills, Knowledge Center moves into Context Center
Breaking. Affects any client, script, or integration calling/v1/agents/abilities or
/v1/knowledgeCenter.
All fifteen /v1/agents/abilities endpoints move to /v1/agents/skills — a pure path rename, the
request and response shapes are unchanged:
Additive. AI Studio itself also gains in-UI authoring for agents, personas, and skills in 2.0,
including drafting a planner description for a dynamic agent. The UI-selected persona is forwarded
to the AI platform, so agent behaviour follows the persona.
Dynamic agents that pointed at the deleted agent-app bots are repointed by the migration.
Separately, Knowledge Center is removed because the capability moved into the platform’s own Context
Center:
The memory endpoints (
/v1/contextCenter/memories/**, 10 endpoints) keep their paths but are now
served by the platform implementation, and three operations are gone entirely: check-duplicates,
promote-chat and remember. Any client calling those three must be reworked, not just repointed.
Context memories are also now visibility-scoped in both AI answers and semantic search
(behavioural) — you see org-wide memories, memories you own, and memories shared with you or
your team, not everything. No action is required; if a memory seems to have disappeared, check who
it’s shared with. /v1/ai/contextPlugins is new and additive: it lets the assistant read from
external sources (Notion, Slack, GitHub, Google Drive ship seeded).
AI Dashboard visibility is enforced, and a chart now belongs to one dashboard
Breaking, and a security fix. Affects every existing AI Dashboard and any client creating AI charts. Dashboard visibility (Private, Shared, Public) was stored in 1.13 but never applied — any user with view rights could open any dashboard. 2.0 enforces it, and new dashboards default to Private. To preserve existing access on upgrade, every existing dashboard is set to Public; owners should review and restrict theirs afterward. Separately, a chart carried a list of parent dashboards in 1.13, and its read access was the union of them — so any user could read someone else’s chart by pinning it to a dashboard of their own. In 2.0 a chart belongs to exactly one dashboard:dashboards: [...] or the
old queryConfig shape must be rewritten — dashboard (singular) and queryType are now required.
Two related endpoint removals: GET /v1/ai/charts (list) is gone — chart listing now hangs off the
owning dashboard at GET /v1/ai/dashboards/{id}/charts — and POST /v1/ai/charts/preview is replaced
by a single consolidated preview endpoint. Scheduled chart refresh is also re-platformed onto in-JVM
Quartz (behavioural): the ingestion pipeline, Argo workflow and Python path that used to run it
are retired, so any monitoring built on those pipelines (alerting on a failed refresh pipeline, for
example) will see the pipelines disappear — monitor the automation and dashboard endpoints instead.
AI Analytics also gains further additive changes in 2.0: