Twym Fabric
Sign in
Continue with Microsoft to access Twym Fabric.
Guided Setup
One place to create a project, add agents (new or reuse), optionally set a team, permission policy, and governance, then get the API endpoint to integrate on your platform.
- 1 Project *
- 2 Agents *
- 3 Team optional
- 4 Policies optional
- 5 API endpoint *
* required · steps without a star can be skipped
Project
You must pick a project before anything else. Create new starts a blank project. Use existing continues with one you already have.
Creates a new project workspace and issues a publish API key.
Select a project you already created. You will add/reuse agents on it in the next step.
Agents
At least one agent must be on the project. Create new builds an agent from a template or your own config. Reuse existing maps an agent already in the registry.
Pick a template and create. Name, description, runtime, model, and tools come from the template. The system prompt is generated from its description below.
Build a specialist agent yourself (does the work when called). The project orchestrator is created on the Team step when you have multiple teams.
System prompt
Generated from the description — the instructions the agent follows when it runs. Preview and edit here, or let it be generated on create.
Pick an agent that already exists. The same agent can be reused across projects.
Agents on this project
Agents added here are always specialists — they answer requests. A supervisor is different: it routes work to specialists and is created/assigned on the next optional Team step.
Team optional
Group mapped agents under a team. Choose Parallel or Sequential
running mode (works with or without a supervisor). Call the project execute URL without
agentId to run the team flow. Click Next if you only need agents on the project.
Map at least two agents to this project before creating a team.
Project orchestrator
Required when you have multiple teams. Fill in details to create a new dedicated routing agent — not one of your specialists.
Policies optional
Choose how agents may act (permission policy) and compliance rules (governance profile). Click Next to keep defaults.
Controls when tool/actions are auto-allowed vs require your approval during execution. New projects get the default policy unless you change it here.
Profiles define allow / deny / redact rules, rate limits, and semantic checks at project level.
API endpoint
Final step — copy these details into your platform.
tf_… key, then Save & connect. Without Atlas
connected, knowledge search and citations will not work on this endpoint.
REST example
Agents
Lists agents in the registry with status, runtime, and model. Search or filter, then open an agent to manage it. Governance companion agents are kept in the background and are not listed here.
| Name | Status | Agent ID | Runtime | Model | Updated |
|---|
Create Agent
Start from a template (config comes preset) or switch to Custom config for a specialist you define yourself. Fields marked * are required.
Agent Details
Shows this agent’s configuration, governance profile, lifecycle actions, and feedback history.
—
Edit Agent
Edits this agent’s name, system prompt, runtime, tools, data sources, and governance profile. Use Agent Details to activate, disable, or archive.
—
Query Tester
Runs a live execute call on an active agent. Shows the response, token usage, governance decision, and a form to submit feedback.
Execution result
Answer
—
Run details
Raw response JSON
Governance JSON
Governance
Create and review governance profiles that agents use at execute time. Each profile holds who can call (roles), rate limits, exact pattern redaction/blocks, and custom policy checks evaluated for meaning (not only exact words). Attach a profile on Create or Edit Agent.
Profile catalog
Profiles already available to attach to agents, including which agents use each one and their custom policy checks.
Permission Policies
Build copyable permission policies from draggable blocks — time, resource, action, milestone, and security. Enable a policy when creating a project (or agent), change it anytime in settings. During execution the Governor sees what the agent is trying to do and enforces the policy, asking the user only when a rule requires a decision.
Logs
Execution history by project. Each row is one run: which agent executed, when, whether it succeeded, how long it took, and how many tokens it used. Open a row for the request, the failure reason, and per-agent results.
Projects
Projects are parent containers for agents. After create you can
Publish (give others your endpoint + fab_… key) and
Connect (attach another Fabric project or external API/MCP with their credentials).
Project registry
Project
Project status
—
What to do next
Teams
Optional — group project agents to run together. Set Parallel or Sequential running mode. Call execute without agentId to run the team flow; pass agentId to run one agent alone.
Map at least two agents before creating a team.
Project orchestrator
Required for multi-team projects. Create a new dedicated orchestrator — do not reuse specialist agents.
Project agents
Mapped specialists on this project. External callers use the project execute URL — not a per-agent URL.
Result
—
Technical result
How other systems can call this project
Changes save automatically. Enable the interface your caller understands.
REST example
payload is whatever your agent’s prompt expects. Leave it {} if the agent needs no input.
Connect — attach other projects / sources
Use this only when agents in this project must call another project or external service.
Retire or delete project
Archive keeps the project on record but stops execution and publishing, so it can be brought back. Delete removes it for good, along with its teams, agent mappings, publish keys and connections — the agents themselves stay in your catalog. Use Disable from Overview if you only need a temporary pause.
Knowledge · TwymAtlas
TwymAtlas is Fabric’s only knowledge plane (MCP). Fabric does not host generic or local knowledge sources — agents retrieve indexed enterprise knowledge from Atlas at execute time.
How to use Atlas from Fabric
Fabric talks to TwymAtlas over MCP — paste your key in the form below. No .env edits or restart required.
-
Connect Atlas
Sign in to TwymAtlas → sidebar MCP → Configure (first key) or Reconfigure (rotate; old key stops immediately). Copy the MCP URL and
tf_…key (shown once). Paste both below and click Save & connect. Open Atlas → MCP - MCP URL origin If the Atlas UI is on a different host than the MCP endpoint, paste the MCP URL from the Atlas MCP page — not the UI origin alone.
- Rotate or revoke keys After Reconfigure or Revoke on Atlas MCP, paste the new key here and save. To remove Fabric’s saved copy, use Revoke console key below.
-
Knowledge agents
Use a knowledge agent (seeded:
agt_ops_knowledge) or any agent with Atlas knowledge / toolknowledge_search. At execute time the agent receives every advertised Atlas MCP tool — you do not list those names in Allowed Tools. -
Via project
Open Twym Ops → map a knowledge agent → Execute (pass
agentIdif multiple agents) with task JSON, or a query field for knowledge assistants. - Direct test Query Tester on Ops Knowledge Specialist, or use the search form below.
-
Citations
When Atlas returns chunks, execute responses include
citations/knowledgeUsed: true.
Connection status
Exposed tools
Tools advertised by TwymAtlas MCP for this API key. Knowledge-capable agents invoke whichever of these they need at execute time — Fabric does not hard-code the list. New Atlas tools appear here and become callable on the next run.
Answer
Route not found
That AI Services path is not configured.