AI teammate · task tracker · any model
One chat with your AI teammate. It turns into tasks on its own.
Talk to pendlooq the way you talk to a colleague in Slack. It turns the conversation into tasks, files them into streams of work and walks each one through its statuses to a result.
Free and local with your own model key. The cloud is optional.
Why pendlooq
Built for people who run many things at once
pendlooq is an open-source AI teammate that turns one chat into tasks, works with any model and keeps your keys on your machine. Team leads, CTOs, managers and independent developers already lean on AI. With dozens of parallel topics, ordinary AI chats become a pile: what is in progress, what waits for an answer, what was dropped? pendlooq is organized as task → result, not chat → history.
The chat is the task tracker
You never file a task separately. The router turns a message into a task, and an Undo chip stays under it in case it guessed wrong.
No duplicate tasks
A message that belongs to an existing task finds it, across all streams and statuses, even archived, and continues it instead of spawning a new one.
Any model
Anthropic, OpenAI, hundreds of models through OpenRouter, or your ChatGPT subscription instead of a key. No lock-in to one vendor, and each stream can use its own model.
Credentials scoped per stream
The assistant in your Personal stream never sees work keys. The model only sees a credential’s name, never its value.
Respects your setup
If a folder already has a Claude Code, Codex or OpenClaw config, pendlooq works with it instead of imposing its own.
Tasks move without you
In the cloud, a task continues on its own when a colleague replies in Slack, and dropped work resurfaces in your digest.
How it works
From a message to Done in five steps
Write the way you would to a colleague. Statuses tell you whose turn it is.
- 01
You write in the chat
“We need to raise the billing limits.” No form, no project picker.
- 02New
The router makes a task
A cheap service model decides: an existing task, a new one, or just a remark. A chip shows the task and its stream, with Undo and Change stream.
- 03Needs me
The assistant asks
It works with the stream’s folders, skills and connectors. When it needs you, the task moves to Needs me.
- 04In Progress
You answer
Reply in the thread. The task goes back to In Progress and the assistant carries on.
- 05Done
Review, then Done
The result lands in Review. Accept it or send it back for changes. The feed records what was done.
Three views, one set of tasks
A chat that becomes a tracker
Switch with Alt+1, Alt+2, Alt+3. Every view opens the same task thread.
Chat
Alt+1The general chat in the middle, streams and their tasks on the left, the open thread on the right. Messages that became tasks fold into one line with a chip.
Board
Alt+2New → In Progress → Needs me → Review → Done, filtered by stream. Start, stop, resume, accept or archive a task by hand.
Feed
Alt+3One line for every action the assistant took, including what it learned and wrote to a stream’s notes.
Streams
Every line of work gets its own space
A stream is a direction of work: billing, hiring, personal. The assistant in a stream works only with what that stream was given.
Folders
Files are available only inside the stream’s folders.
Skills
Instructions in the open SKILL.md format, attached per stream.
Connectors
Jira, Notion and Google Drive, each with read, write-with-confirmation or write access per stream. Slack in the cloud.
Credentials
Passwords and tokens live in your OS keychain. The model sees a name, never the value.
Notes and settings
Long-term memory and agent permissions in familiar formats: Claude settings and AGENTS.md, or the Claude Code, Codex or OpenClaw config already in the folder.
- Folders
- ~/work/billing-service
- Skills
- code-review, release-notes
- Connectors
- Jira · write with confirmation
- Credentials
- BILLING_API_TOKEN
- Model
- per stream or default
One-off question? It lands in “No stream”: the assistant answers and reasons, without files or connectors. Move it to a stream later.
Any model
Bring the model you already pay for
pendlooq is not tied to one vendor. Connect a key, pick a default model, and give the router and the feed a cheaper service model.
Anthropic
API key
OpenAI
API key
OpenRouter
One key, hundreds of models
ChatGPT
Sign in with your subscription
- Each stream can override the default model.
- Keys are stored in your operating system’s keychain.
- Local models via Ollama are planned after the first release.
Your keys stay with you
The cloud coordinates. Your runner executes.
The agent runs on your machine: your laptop by default, or your own server. Model keys, credentials and files never reach pendlooq servers, and the cloud never calls a model.
Any device
- Browser
- Phone
pendlooq cloud
- Chat, tasks and feed
- Queues and notifications
- Slack thread tracking
Your runner
laptop or server
- Model keys and credentials
- Stream folders and files
- Runs the agent and its tools
Model provider
- Called with your key
What the cloud adds
The local app works while your computer is open. The cloud keeps going.
Tasks move on their own
The assistant asks a colleague in Slack and the task waits in Needs others. When the reply arrives, the task continues without you.
“Your turn” notifications
Know the moment a task needs your answer or is ready for review: in the browser, by email or in Slack.
Daily or weekly digest
What is stuck, what waits for an answer, what was auto-archived. Built from your task list, without a model.
Any device
Answer a question, accept a result or start a task from any browser. New tasks wait until your runner is online.
Open source
Read the code you trust with your keys
You hand pendlooq passwords, tokens, folders and conversations. That is easier to do when you can read the code, especially the runner, which holds the keys and runs commands.
Client and runner are open
Interface, local API and runner are published under the Functional Source License (FSL-1.1-ALv2).
Apache 2.0 after two years
Every version converts to Apache 2.0 two years after release.
The cloud is closed
Sync, thread tracking, notifications and billing are proprietary. That is what the subscription pays for.
Pricing
Free locally. Pay only for the cloud.
You pay your model provider directly. pendlooq charges for the cloud, never for tokens.
Free
The whole local app with your own model key.
$0forever
Your keys, your machine
Pro
CloudThe cloud: tasks that move on their own, on any device.
$—per month
Pricing at launch
Team
For teams that share lines of work.
Soon
Per seat, priced later
Coming after Pro
FAQ
Questions, answered
Do I pay for tokens?
No. You connect your own key or provider subscription and pay the provider directly. pendlooq charges for the cloud and never sells tokens or credits.
Where do my tasks run?
On your computer or your own server, in the pendlooq runner. The cloud stores, coordinates, queues and notifies; it never calls a model. A task runs whenever the runner of its stream is online.
What happens if I cancel?
Pro stays active until the end of the paid period. The local app keeps working. Your cloud data stays available to download or move back to your computer with one click.
Do you train models on my data?
No. pendlooq does not train models on your data, and the cloud never sends your data to a model: the runner calls your provider with your own key.
Which models can I use?
Anthropic and OpenAI with an API key, hundreds of models through OpenRouter, or your ChatGPT Plus or Pro plan through Sign in with ChatGPT, without an API key. Each stream can use its own model.
Is pendlooq open source?
The client and the runner are source-available under the Functional Source License; each version becomes Apache 2.0 after two years. The cloud is proprietary.