Documentation Index
Fetch the complete documentation index at: https://docs.phrony.com/llms.txt
Use this file to discover all available pages before exploring further.
A Phrony manifest is a YAML document (or stream of documents) that declares resources in your workspace: LLM providers, integrations and their operations, agents, agent versions, and triggers. Use it to back up setup, move work between environments, or review changes before you apply them.
The product speaks the same model as the dashboard. Applying a manifest reconciles those declarations with what already exists—creating, updating, or aligning records so your workspace matches the file.
Where you use manifests in Phrony
| Action | Where |
|---|
| Export manifest | On an agent’s detail page, or from the agents list (per agent). Downloads a YAML file (for example phrony-manifest-<slug>.yaml). |
| Apply manifest | Import manifest on the agents list, or Apply manifest on an agent detail page. Upload YAML; optionally run a dry run first to preview planned changes without writing. |
Document shape (version 1)
Every manifest document must include:
| Field | Value |
|---|
kind | phrony.manifest (legacy exports may use phrony.workspace.manifest; both are accepted.) |
version | 1 |
| Section | Purpose |
|---|
metadata | Export hints: optional exportedAt, label, rootManifestKey. |
llmProviders | Named model backends referenced by agents. |
services | Integrations (http or built_in) and their operations. |
agents | Agent identity: manifestKey, name, executionMode, llmProviderName. |
versions | Per-agent version snapshots (model, instructions, tools, limits, …). |
triggers | How work enters agents (schedule, API, or event). |
| Field | Description |
|---|
exportedAt | ISO timestamp from export (informational). |
label | Optional human note. |
rootManifestKey | Optional logical key of the “root” agent in a multi-agent export. |
LLM providers (llmProviders)
Each item is a named provider your agents can use:
| Field | Description |
|---|
name | Unique name; agents reference this in llmProviderName. |
type | One of: openai, openai_azure, anthropic, self_hosted. |
baseUrlOverride | Optional override URL for the vendor API (often null or omitted). |
Services and operations (services)
Each item is an integration-shaped record:
| Field | Description |
|---|
manifestKey | Optional stable key for reconciliation; if omitted, identity falls back to name. |
name | Display name; referenced by operations and by allowedOperations on versions. |
type | http or built_in. |
config | Connection settings (base URL, auth shape, built-in toolkit slug, defaults, …). Sensitive pieces may be redacted on export (secretsRedacted may be set). |
secretsRedacted | When true, the export removed secrets; you must re-supply connection details or secrets where needed. |
operations | Optional list of tools exposed from this integration (see below). |
| Field | Description |
|---|
manifestKey | Optional stable key for this operation. |
name | Operation name; paired with serviceName in allowedOperations. |
description | Optional text for operators or tooling. |
configuration | HTTP path/method/headers, or built-in tool slug and version—same concepts as the operation editor in the dashboard. |
requireApproval | Whether this tool requires approval when your execution mode allows it. |
argumentsSchema | Optional JSON Schema for tool arguments. |
secretsRedacted | Mirrors service-level redaction for this operation if applicable. |
Agents (agents)
| Field | Description |
|---|
manifestKey | Required stable id used everywhere else in the file (agentManifestKey, allowedSubAgents, triggers, …). |
name | Human-readable name. |
executionMode | request, hitl, or sub_agent (same semantics as Agent). |
llmProviderName | Must match a llmProviders[].name. |
Agent versions (versions)
Each item is one version snapshot for an agent identified by agentManifestKey:
| Field | Description |
|---|
agentManifestKey | Links to agents[].manifestKey. |
status | testing or deployed. |
versionLabel | Label for this line (for example v1). |
llmModel | Model id for the chosen provider. |
instructions | Optional system-style instructions. |
description | Optional short description (length limits apply in product). |
inputSchema / outputSchema | Optional JSON Schema objects. |
temperature | Optional sampling temperature. |
maxIterations, maxTokensPerRun, maxToolCalls | Optional run budgets. |
anomalyControl | Optional L1 anomaly behavior (see Guard overview and L1 rules). |
canExecuteSubAgents | Whether this agent may delegate to sub-agents. |
subAgentExecutionModel | sequential or parallel when sub-agents are allowed. |
maxSessionTokens, maxSessionDurationSec, topP, topK | Optional session and sampling limits. |
rules | Optional list (product-specific rule objects). |
allowedOperations | Optional list of { serviceName, operationName } entries that become allowed tools. |
allowedSubAgents | Optional list of sub-agent manifestKey values this version may call. |
allowedOperations uses integration name and operation name as shown in the manifest’s services section, not arbitrary labels.
Triggers (triggers)
Each trigger targets one agent (agentManifestKey) and is one of:
Scheduled
| Field | Description |
|---|
name | Trigger name. |
type | scheduled |
cronExpression | Cron string. |
timezone | IANA timezone name. |
inputPayload | Optional JSON object passed into runs. |
manifestKey | Optional stable key. |
| Field | Description |
|---|
name | Trigger name. |
type | api |
rpmLimit, rpdLimit, maxConcurrentRuns, maxInputTokens | Optional rate and concurrency limits. |
ipWhitelist | Optional list of allowed callers. |
exposeStepTimelineToApi | Optional visibility flag for step timeline in API responses. |
manifestKey | Optional stable key. |
| Field | Description |
|---|
name | Trigger name. |
type | event |
serviceName | Integration name the event is tied to. |
eventSlug | Event identifier for that service. |
eventConfig | Optional JSON object. |
manifestKey | Optional stable key. |
Multiple YAML documents and includes
You can put several YAML documents in one file, separated by --- (a multi-document stream). They are merged into a single logical manifest: later values override earlier ones for the same logical keys (for example same manifestKey on an agent or service).
You can also add index documents that only pull in other paths or snippets:
kind: phrony.manifest.index
version: 1
includes:
- ./shared-integrations.yaml
- ./my-agents.yaml
Dry run and apply result
A dry run returns a plan: which resources would be created, updated, left unchanged, or deleted—without persisting changes. Use it to review diffs before applying.
A real apply returns a result that includes the root agent id used for the operation, a list of changes, optional warnings, and—when integrations still need OAuth or API keys—a list of services needing connection so you can finish setup in the dashboard.
Agent
Execution modes, versions, and triggers in the product model.
Integration
HTTP and built-in services and how operations become tools.
LLM provider
Saved vendor connections agents use for model calls.
