Superpower Daily
Cognition Reportedly Seeks $1B at $47B, Three Months After $26B...EU Seeks AI Act Answers From 30-Plus Companies, With Fines for...CrowdStrike’s SafeMind Pits Red and Blue AI Agents in an NVIDIA...Anthropic’s EFS Puts Misuse-Monitoring Data in Customer Clouds,...World Labs Launches Atlas for 1440p Camera-Controlled Video, 3D...Enflame Opens $910M Shanghai IPO Subscription, Testing Its Spec...Cognition Reportedly Seeks $1B at $47B, Three Months After $26B...EU Seeks AI Act Answers From 30-Plus Companies, With Fines for...CrowdStrike’s SafeMind Pits Red and Blue AI Agents in an NVIDIA...Anthropic’s EFS Puts Misuse-Monitoring Data in Customer Clouds,...World Labs Launches Atlas for 1440p Camera-Controlled Video, 3D...Enflame Opens $910M Shanghai IPO Subscription, Testing Its Spec...
Toolspublished

Perplexity Documents Versioned Agent API Profiles for Pinning Stable Configurations

The configuration layer packages a run’s model, instructions and tools under one ID. Teams can pin a known setup, while requests using latest automatically take newly uploaded versions.

By 2 min read
Perplexity Documents Versioned Agent API Profiles for Pinning Stable Configurations
Perplexity Documents Versioned Agent API Profiles for Pinning Stable Configurations
Sign in to save and react

Listen to this story

The audio brief

About 1:30
0:001:30
Read transcript
Perplexity has documented Profiles for its Agent API, giving developers a versioned way to package an agent’s configuration under one ID. That package can include the model, a fallback chain, system instructions, reasoning effort, tools, and the agent-loop step budget. The important choice is what happens when the setup changes. Pin an exact profile version, and the configuration stays unchanged for reproducible runs. Use latest, and each request can pick up a newer version uploaded later. Perplexity resolves that choice when the request is admitted, so an edit made during a run cannot change that job mid-execution. Profiles sit above the Agent API Perplexity introduced in March for workflows combining search, tool execution, and multi-model orchestration. They are meant to give teams control over the configuration itself, rather than just preserving a prompt. A request must choose either a Perplexity-managed preset or a developer-defined profile, never both. Profile values are defaults, so request-level overrides can still change matching settings. Tools are handled differently: overrides merge per tool, meaning changing one does not switch off the others. And invalid setup is rejected before execution. Unsupported models, bad profile IDs, or invalid versions return a 4xx response. The key constraint to watch is operational: choosing latest makes future configuration edits part of the workload automatically, while pinning makes every change an explicit release decision.

Story brief

3 key points

Perplexity is adding Profiles to its Agent API so developers can manage agent configurations as versioned release artifacts instead of inheriting Perplexity’s changing presets. A profile can capture models or fallback chains, instructions, reasoning effort, tools, and loop budgets. Teams can pin an immutable version for reproducibility or use latest to adopt future edits. Request-level overrides preserve...

  1. 01

    Profiles can bundle models, fallback chains, system instructions, reasoning effort, tools, and agent-loop step budgets.

  2. 02

    Requests must choose either a Perplexity-managed preset or a developer-defined profile, never both.

  3. 03

    Pinned versions remain unchanged; latest resolves when a request is admitted and can adopt later uploads.

Perplexity has documented Profiles, a reusable and versioned configuration layer for its Agent API. A profile stores the settings that shape an agent run under one ID, rather than requiring developers to repeat the full setup in each request. The key operational choice is whether to pin an exact version or let a request follow later changes through latest.

Profiles can bundle a model or model fallback chain, system instructions, reasoning effort, tools and an agent-loop step budget. They extend an Agent API introduced in March as a managed runtime for workflows with integrated search, tool execution and multi-model orchestration. The new layer is aimed at saving a configuration developers control, not merely saving a prompt.

Two paths for configuration control

A request must use either a preset or a profile, not both. Presets are maintained by Perplexity and resolve to its current recommended configuration. Profiles are defined and versioned by the developer, offering teams a way to take control of a setup rather than continually inherit a managed default.

One profile, with controlled exceptions

A profile supplies defaults rather than a sealed package. Request-level parameters override matching profile values, allowing a developer to change one setting for a run. Tools are the exception: their settings merge per tool, so changing one tool’s options leaves other profile tools enabled.

Updates become a release decision

Each profile version is immutable: editing a profile creates another version. A pinned request runs its selected configuration, while latest picks up newly uploaded versions. Perplexity resolves the version when a request is admitted, so an edit made during a run does not change that in-flight job.

Profile errors are request errors

  • An unsupported model named in a profile fails before the run begins.
  • An incorrect or inaccessible profile ID fails before the run begins.
  • An invalid profile version fails before the run begins.

These problems return a 4xx status before agent execution starts. That makes version selection and profile access part of ordinary API error handling, while the pin-versus-latest choice determines whether later configuration edits reach a workload automatically.

Sources

  1. docs.perplexity.aiProfiles - Perplexity
  2. community.perplexity.aiIntroducing the Agent API — A Managed Runtime for Agentic Workflows

Related coverage

Continue reading