Explore and edit entities
The Context Graph connects what exists in your engineering organization with what your agents know about it. Services, teams, people, repositories, agents, skills, and memories appear together, so you can follow a service's dependencies, find its owner, and discover the knowledge linked to it.
Open Context Graph → Graph from the sidebar.
The three building blocks
- An entity is one thing in the graph: a service such as
payments-api, a team, a person, or a repository. - A blueprint is an entity type. It defines the properties and relationships an entity can have. For example, the Service blueprint defines a service's description, tier, and owner relationship. See Blueprints.
- A relationship connects two entities. A service depends on another service, is owned by a team, and has contributors. See Relationships.
The built-in entity types include:
| Blueprint (entity type) | What an entity represents | Example relationships |
|---|---|---|
| Service | A software service or application component | Depends on services · owned by a team · has contributors |
| Team | An engineering team or squad | Has members · owns services and repositories |
| Person | A team member or on-call engineer | Belongs to teams |
| Code Repository | A code repository | Owned by a team · part of a service |
Integrations bring their own entity types, such as Slack channels and ServiceNow configuration items. You can define your own blueprints in the UI, from a CSV file, or through the API. See Custom blueprints. Open Context Graph → Setup → Blueprints to see the available types, their entity counts, and their fields and relationships.
Explore the graph
Graph and Table are two views of the same loaded entities:
- Graph shows entities as nodes and relationships as edges. Select a node to open its details and grouped relationships. Solid edges are blueprint-defined relationships; dashed edges are label-derived relationships, created when a label's key is a blueprint and its value names an entity. Choose Focus connections in the details panel to frame its neighborhood.
- Table lists entities with their type, team, labels, sources, relationship count, and last update. Select a row to inspect it. Use the settings button to show, hide, or reorder columns.
Expand the search icon to find an entity by name or reference. In Graph view, use the arrow keys and Enter to select a result.
Open Filters to choose Blueprints, Team, or Source. The active filters appear beside the controls; remove one with its × button. The initial blueprint selection shows services, teams, repositories, skills, and memories, plus environments when available. Clear that selection to include every available type.
The graph's settings button controls Show relationship names and Neighborhood. Show relationship names on hover or selection, always, or hide them. With an entity selected, focus on its direct connections or connections within two hops.
Filters, selection, graph settings, and the table's column layout are saved in the URL. Copy the URL to share a view. A project overview loads at most 2,000 entities and 6,000 connections, with a fair share for each blueprint; the footer indicates when the graph is larger than the loaded view.
Where entities come from
Most entities come from your integrations. GitHub, GitLab, Bitbucket and Azure DevOps bring repositories; PagerDuty, ServiceNow, Datadog and Jira bring services, teams and people; Slack brings people, user groups and channels. The resolver links what each integration says about the same thing to one entity. See Sync integrations for what each integration brings and when it syncs.
Open Context Graph → Setup → Sources to see two groupings:
- Scheduled syncs groups syncs by entity type, with the integrations that supply each type. For example, People combines your organization's users, Slack, PagerDuty and other sources. Sync now runs that sync immediately.
- Connected integrations lists each integration and the entity types it brings. View sync opens the integration's Sync tab, where you see its entities, its last sync and its settings, and run Sync now.
Last change is the newest update to any entity in the row. It includes manual edits, so it can be newer than the last sync.
Each entity's Sources tab lists the integration records linked to it, the value each source supplies for each field, and which value the entity uses. See Field precedence.
Create and edit entities
Use the + button on Graph to choose any available blueprint, including built-in, integration-specific, and organization-defined entity types. The create form uses the selected blueprint's fields.
Select an entity and choose Open page to read its properties, relationships, sources, linked knowledge, and local graph. Choose Edit to change its title, properties, labels, and relationships. A value you edit is declared: it takes priority over every integration, and the next sync does not overwrite it. To hand a field back to the integrations, clear your value.
Each relationship offers targets of the type its blueprint allows. A relationship with one target, such as a service's owner team, replaces the previous target when you select another.
Choose Delete on the entity page to remove it. If other entities reference it, Autoheal names those references and asks for another confirmation before deleting it anyway. Creating, editing, and deleting require the corresponding project permissions.
Connect agents, skills, and memories
A label connects a resource to an entity when its key is a blueprint name and its value exactly matches an entity's canonical name. For example, service: payments-api links a skill to that service. Other label keys describe the resource without creating a connection. See Label associations for the matching rules.
A skill's labels live in its SKILL.md frontmatter. An agent's labels live at the top level of agent.yaml. Both write each label as one - key: value item, and repeat a key for each of its values:
labels:
- team: payments
- service: payments-api
- service: refunds-api
The Labels picker on the skill or agent page opens the file editor with the chosen label added to the draft. Commit the edit to publish it. Labels travel with the file and its branch; they do not grant access or change what an agent runs. Memories, artifacts, and runs keep their labels in Autoheal.
Four label keys name Autoheal resources instead of entities:
| Key | Value | Example |
|---|---|---|
agent | The agent's name, as in its folder | agent: payments-triage |
skill | The skill's name | skill: refund-procedure |
artifact | The artifact's name | artifact: weekly-incident-report |
run | The run's ID | run: 3f2c… |
These keys always name resources, also when a blueprint has the same name. A value that matches nothing is reported on the labeled resource. A key for a resource type that you cannot read is ignored.
Some links need no label: a run is linked to the agent that ran it, and an artifact that an agent made is linked to the run that made it. These links show as label-derived relationships marked as implicit. You cannot remove them by editing labels.
The graph shows the agent runs of the last 30 days, at most 300. You need permission to read runs (tool:read) to see them.
How agents use the Context Graph
During a run, an agent can read the Context Graph to resolve ownership, follow dependencies, find the right repository, and discover linked skills and memories. It can also write what it learns, such as a dependency it found in traces. Its reads and writes remain subject to the run's permissions and its context_graph scope. See Agent tools.
Review suggestions
When the resolver is not sure that two records describe the same thing, or an agent's value conflicts with an integration's, it creates a suggestion instead of changing the graph. Open Context Graph → Review to accept or reject each one. See The review queue.