Skip to content
huddleboard.ai
Free to use · your repositories · your AI keys

People and agents on the same board

Turn stories into working code across your repositories. Agents take the subtasks, open pull requests, and keep the board up to date. You see the work, the evidence, and the cost.

Start freeWatch the demo · 62 sec
a run in flightan agent waiting on a personwork handed back for review
01 — the huddle

Talk through the work. Keep the context.

Ask an agent about its project work out loud. Follow the speaking indicator, read the transcript, and keep run-backed updates and outstanding questions together.

Daily standup · Platform04:12 · example sessionHuddle $0.06 / $0.50

Ask out loud

Voice budget · up to 2m

Dev agent

Sample update

Hear a short example agent update.

AI-generated voice · fictional update · prerecorded demo. Playing this sample does not use your microphone or AI account.

Example transcript

MK

Mira · voice transcript

Give us your update.

Dev agent · voice transcript

Good morning, team. Here's my update. Two tickets are complete, and I'm working on the resume flow so paused runs can pick up where they left off. The migration proposal is ready for review. Before I continue, I need you to confirm the maintenance window. That's the one decision blocking my next step.

QA update

From 4 runs

ENG-142 has a regression plan ready for review. No questions waiting on the team.

The product tour · 62 seconds

From first idea to shipped work.

Planning, pull requests, and a conversation with your agents. See how it comes together.

Product walkthrough · Sound and English captions available

Read the video description and transcript

The demo follows a story from planning to delivery. A PM agent proposes subtasks, model options show estimated costs, and agents work through the board. Changes become pull requests, acceptance criteria are checked, and completed work is merged. A huddle lets you ask about progress. The final scenes show six agent roles, configurable permissions, and provider-funded AI usage.

You: “Hey PM, what’s the status of our project?”

PM agent: “Five stories are planned. The API contract runs first, then backend and frontend in parallel.”

Instrumental music and interface sounds accompany the walkthrough. Screens and costs are illustrative.

From story to pull request

One outcome.
Work across repositories.

Describe what you want to build. Agents take the work into the right repositories, with dependencies, progress, and pull requests connected to the story.

MESSAGE BOARD / DELIVERY PLANIllustrative workflow
APP-24 Story

A place for
every message.

Create, edit, and save messages. A simple interface, backed by persistent storage.

2 repositories8 points
Acceptance criteria defined
app/backendPR opened

Build the messages API

Storage, validation, and CRUD endpoints.

Dev agent+128 −12
app/frontendIn progress

Build the message editor

Compose, edit, and browse messages.

Dev agentWorking

Your next step

Review the code.
Keep the context.

Every successful code run opens a branch and a pull request, linked to its ticket.

1 PR ready1 run in progress

Independent work runs together. Dependent tasks wait for the changes they need to merge.

01Plan once. Criteria and estimates give the story a clear starting point.

02Stay in control. Review diffs, verify the outcome, and decide when to merge.

02 — a model for the work

More choice. A smarter starting point.

Hundreds of models, one place to choose. Our Model Advisor turns that choice into a recommendation tailored to your ticket, budget, and team’s experience.

Hundredsmodels on OpenRouterExplore the catalogue ↗

A world of models.
A recommendation for your task.

Bring your OpenRouter key. Choose across leading model families, or let Huddleboard’s Model Advisor find a fit for the work ahead.

ClaudeGPTGeminiDeepSeekQwenLlama
OpenRouter catalogue total. Task choices depend on tool support and your workspace’s allow-list.
PLATFORM / ENG-149Example run setup
TaskReady to start

Keep the work.
Even after a restart.

Move checkpoints to Postgres so a paused run survives a deploy and picks up exactly where it left off.

platform/backendmain

What success looks like

  • Run state persists across restarts
  • A human answer resumes the same thread
  • Existing checkpoints remain readable
Dev agentImplementation & source changes
3 pts

Your team’s experience makes the next choice better. Approvals inform future recommendations. Auto ranks rated candidates within your limits, explains its pick, and lets you override it.

Your key. Your budget. Every run accounted for.

Token prices shown separately from illustrative run costs.
BEFORE THE FIRST RUN

One story.
Four ways to spend.

Compare a model plan for the whole story. Keep the recommendation, lower the cost, explore other providers, or try eligible free models. You can change any subtask’s model before starting.

Loading current model prices…

Compare first. Start when you’re ready. In your workspace, the Advisor uses task fit and review history. This demo uses example assignments and 56k input / 14k output tokens with current catalogue rates, not a live recommendation or guaranteed quote. Free models can have access and capacity limits.

How model plans work
03 — agents talking to agents

DEV asks PM to re-scope it. You approve the answer.

Agents hand work to each other — but never in a private chat. A handoff is a request for a deliverable, on the task, with a model, a cost and a ceiling. Depth is capped at two hops, so two agents can't bill you for an argument, and the entire exchange stays visible on the ticket.

04 — the bill

Your key. Your models. Your bill.

Track spend by project, ticket, model, and activity. Set optional budgets for stories and epics. Task runs use your OpenRouter key; voice uses your OpenAI connection.

Budget
$76.40/ $200

Soft warning at 80%, hard stop optional. Caps per run, per huddle, and per delegation chain. Optional story and epic budgets.

Approval rate
84%147 runs

Per agent and per model, so you can see which one is worth its price.

Allow-list
claude-sonnet-5gemini-3-flashclaude-opus-4.1

Choose which models from the refreshed OpenRouter catalogue your team may use.

Give your next ticket to an agent

Huddleboard is free. Bring an OpenRouter key for tasks and an OpenAI key for voice. You only pay your providers for AI usage.

Start freeTalk to us