Aster

Aster by AsterWise is an LLM router with one OpenAI-compatible API and two choices: Aster Work for writing, research, and everyday tasks, or Aster Code for coding and repository work. Agents send a prompt or a full chat completion with history, function tools, and reasoning effort, and Aster selects the underlying model for each request.

Verified 2 actions API key available

Platform Documentation Capability data verified 2026-10-05

A new issue lands in the repository and the triage flow picks it up

Agent pulls the issue body, the stack trace, and the source files the trace names

Agent confirms the trace points at real paths in the repo before spending a request

Agent calls Generate Text against Aster Code with the issue and the surrounding source

Agent reads the finish reason and discards an answer the token limit cut off

Agent posts the proposed change to the engineering channel with the model Aster reports and the token usage

A maintainer approves the comment before the agent posts it on the issue

What This Integration Enables

Aster by AsterWise takes model selection off the table. Instead of a catalog to shop in, it offers two choices: Aster Work for writing, research, and everyday tasks, and Aster Code for coding and repository work. You declare the kind of work; Aster decides which underlying model serves the request, and keeps deciding as models change. For a team running automation across dozens of flows, that is the useful trade. The choice you encode in a flow is the one that stays true, because it describes the job rather than naming a model that will be superseded.

Two actions cover the surface. Generate Text sends a single prompt with optional system instructions and returns plain text alongside the finish reason, the token usage, the model name Aster reports, and the completion ID. Create Chat Completion runs the full OpenAI-compatible shape: conversation history, function tools, tool choice, parallel tool calls, reasoning effort, an output token limit, and sampling controls. Aster returns tool requests but never executes them, which means the tool loop lives in your flow where you can see it, and a human-in-the-loop step can answer a tool request just as easily as an API call can. The connector validates the model's limits before anything leaves FlowRunner: reasoning effort, tool choice, and the output ceiling are checked against the chosen model, so an unsupported combination is refused instead of billed. Output ceilings differ by door, up to 32,768 tokens on Aster Work and 65,536 on Aster Code, and responses come back in one piece because this connector does not support streaming. Aster has no triggers. Nothing here listens for events, so every flow that uses Aster starts somewhere else: a schedule, an inbound record, or another service's trigger.

Without FlowRunner

Model choice by committee Every new flow reopens the argument about which model to point it at

Defaults that quietly age A model named in a flow last quarter is still named there after a better one ships

Truncation read as completion An answer cut off at the token limit looks finished and moves downstream unchecked

With FlowRunner

Two doors instead of a menu The flow picks the kind of work, Aster Work or Aster Code, and Aster picks the model

Routing changes in one place New models arrive behind the door, not across every flow that calls it

Finish reason checked, not assumed The agent reads why generation stopped before anything acts on the text

Use Case Scenarios

Issue triage that proposes a fix

A new issue opens in GitHub with a stack trace and no reproduction steps. The agent reads the trace, resolves the files it names, and calls Generate Text against Aster Code with the issue text and the relevant source. It checks the finish reason before using the answer, because a diagnosis cut off mid-sentence is worse than no diagnosis. The proposed explanation goes to the engineering channel with the model Aster reports and the token usage attached. A maintainer approves it, and the agent posts the comment on the issue. The triage queue gets a first pass; the repository gets nothing a person did not read.
Inbound email drafted, not sent

A message arrives in a shared mailbox through Gmail. The agent classifies it, pulls the relevant account history from the systems that hold it, and calls Generate Text against Aster Work to draft a reply in the account's own context. The draft lands in front of the owner with the source material beside it. They send, edit, or discard. Aster Work handles the blank page, which is the slow part. The account owner keeps the relationship, which is the part that carries risk.
A tool-calling agent with a supervised loop

A support thread needs an answer that depends on live data. The agent calls Create Chat Completion with the conversation history and a set of function tools, and Aster answers with the tool calls it wants rather than running them. Each request becomes a step in the flow: a read goes straight through, and a write stops for a person first. When the loop finishes, the agent appends the tool results, asks for a final answer, and posts the summary to Slack. The model gets to ask for whatever it needs. What it actually gets to do is a decision the flow makes, not one the model makes.

Human-in-Loop Highlight

Successful Aster requests are charged against your prepaid balance. Failed and cancelled requests are not charged. A client-side timeout does not establish final status or charge; check Usage before retrying. That timeout is the one case the connector cannot settle on its own. When a call aborts on its timeout, the original may still be processing on Aster's side and may still complete and be charged, so a blind retry can pay twice for work that already happened once. There is no usage or balance action in this connector to settle the question; that record lives in the Aster app. So the agent pulls the digital andon cord instead of guessing. It stops the flow and asks the operator: "The Aster Code request for issue 4471 aborted after 120 seconds. It may still be processing and may already be charged. Check Usage in the Aster app, then tell me: retry as is, raise the timeout and retry, or stop here?" A retry is cheap to authorize and expensive to automate, which is exactly the shape of decision that belongs to a person.

Agent processes routinely

Detects exception requiring judgment

Clear match Continues automatically

Ambiguous Routes to human via preferred channel

Human decides

Agent resumes with decision

Agent Capabilities

2 actions

Text Generation

Generate Text Sends one prompt to Aster, with optional system instructions, and returns the generated answer as plain text together with the finish reason, token usage, the model name Aster reports, and the completion ID. Used for the single-shot jobs that make up most of the work: draft a reply, summarize a thread, explain an error, name the likely cause of a failure. A finish reason of length means the token limit cut the answer off, which is worth checking before anything downstream treats it as complete.
Create Chat Completion Runs a full OpenAI-compatible chat completion: the conversation history as messages, optionally with function tools, a tool choice, parallel tool calls, reasoning effort, an output token limit, and sampling controls. Returns the complete Aster response plus the first choice's text, its tool calls, the finish reason, and the usage. Used when the exchange has state, when the model needs to ask for data before it can answer, or when reasoning effort should be turned up for a harder question. Aster returns tool requests and leaves the running to you, so each tool becomes a flow step with its own guards.

Frequently asked questions

What can FlowRunner do with Aster?

FlowRunner agents can run Generate Text and Create Chat Completion in Aster.

Does connecting Aster to FlowRunner require OAuth?

No. Aster connects to FlowRunner with an API key, no OAuth flow required.

Can Aster trigger a FlowRunner workflow automatically?

Aster doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.

Start building with Aster

Free plan, no card required. Connect in minutes.

Start Building Free Talk to us