Chat
Work with governed Spark agents that can investigate pipelines, ask connected data, explain notifications, and report through approved channels.
Caller RBAC · per-agent MCP allow-list · persisted conversations and tool tracesAI Studio · Gluesync 2.3
Build agents. Run them like a platform.
Chat is only the first workspace. Publish immutable Spark agents, route them across customer-managed models, run durable jobs, and govern every application call from inside Core Hub.
AI Studio
Core Hub · 2.3 Spark agent
Investigate billing lag
Ask about pipelines and data with the caller’s Core Hub permissions.
Live evidence
3 tool calls Ready
Read-only Verified
Trace saved Verified
Agent v4
Publish an immutable version
Combine instructions, versioned skills, routing policy, and an output contract.
Live evidence
2 skills Ready
TEXT Verified
Content hash Verified
Run 8F12
Follow durable execution
Track queued, routing, and running events without tying work to a browser session.
Live evidence
SUCCEEDED Ready
18 events Verified
Idempotent Verified
Routing preview
See why a model wins
Preview policy decisions against capability, residency, health, price, and cost limits.
Live evidence
1 selected Ready
3 fallbacks Verified
No tokens spent Verified
Platform access
Bound automation before it runs
Issue scoped API keys, assign budgets, and inspect usage and audit evidence.
Live evidence
gsa_•••• Ready
2 scopes Verified
$42 budget Verified
One studio · Five workspaces
From the first prompt to repeatable production work.
AI Studio separates exploration, authoring, execution, model policy, and governance so a useful chat can become a controlled platform asset. Meet Spark, the agent layer running inside it.
Build & publish
Turn instructions and reusable skills into immutable, versioned agents with input, output, routing, duration, and cost boundaries.
Versioned skills · content hashes · JSON output contracts · Craft with SparkRuns
Start asynchronous work and retain its state, selected model, redacted errors, output, correlation, and ordered event timeline in Core Hub.
Idempotency · cancellation · SSE events · ambiguous-outcome disciplineModels & routing
Manage a canonical model catalog and preview deterministic policy decisions before a real invocation spends tokens.
Capabilities · residency · health · price · latency · ranked fallbacksGovernance & developer API
Control programmatic access and estimated spend, then inspect platform usage and audit evidence from the same workspace.
Budgets · scoped gsa_ keys · rate limits · owner and model allow-listsThe Core Hub foundation
Your models. Gluesync’s operational context.
The provider can change. The identity, tools, policies, and evidence around the work stay consistent.
Read the AI Studio documentation →Bring your own model estate
Connect OpenAI, OpenAI-compatible endpoints, Anthropic, Azure OpenAI, or local Ollama. Credentials remain encrypted in Core Hub and model traffic never proxies through MOLO17.
One Enterprise brain
Spark agents ground every turn in the Enterprise brain: the schema graph of every connected database, memory shared across the team, published skills, and approved company tools. External MCP clients such as Claude, ChatGPT, Grok, or Meta Muse reach the same brain through the 125-tool Core Hub MCP catalog, under the caller's permissions.
Explore →Policies before provider calls
Filter models by capability, provider, context, residency, health, price, and estimated cost. Routing preview explains every inclusion and exclusion without consuming tokens.
Evidence after every run
Conversation traces, immutable versions, event timelines, usage records, and safe audit details make answers and automation reviewable without leaking provider secrets.
From chat to workflow
A published agent version is the unit Chronos schedules, platform events, webhooks, chained steps, and your applications fire. AI Workflows add payload allow-lists, idempotency, loop guards, budgets, and approvals around every unattended run.
Explore →Proven use cases
Ask your data, triage sync health, explain alerts, discover PII, stand up pipelines, validate migrations, and embed agents in applications: each with the prompt that opens it and the tools it needs.
Explore →Developer platform
Put published agents behind the interfaces your applications already use.
POST /api/ai/v1/agents/{slug}/runs GET /api/ai/v1/runs/{id}/events POST /v1/chat/completions Use the native run API for durable execution and event history, or point an existing OpenAI SDK at Core Hub for text completions. Scoped platform keys limit agents, models, requests, expiry, and spend.
2.3 runtime boundary: published runs and OpenAI-compatible completions are TEXT-only. Tool-enabled published runs, run approvals, and provider token streaming are not available in this runtime.
Your infrastructure · your models · your policies