CharClaw
Runs on a Machine You Control
Agents run on your own machine or a server you choose, so the work stays next to your environment and your tools.
AI Coding Agents on Your Issue Board, Working on a Machine You Control
Assign a GitHub issue to an AI coding agent, which codes, commits and pushes on your own machine with your own keys. Book a demo and we will walk you through it.
Our Point of View
An AI coding agent belongs where the code already lives, answering to your issue board and your review, so that its work stays easy to inspect and easy to govern.
The Situation
Running a coding agent from a terminal is simple enough for one developer and one task. Once a team is involved, people need to know who asked for what, which agent is working on it and what it changed, and they need the work to stay close to their own environment, tools and credentials.
Our Approach
CharClaw treats agents as members of the team. You assign a GitHub issue to an agent, and it works on a branch on a machine you control, using your own keys, while its progress streams back to the board. Your reviewers then decide what is merged, as they do now, and each completed session leaves a reusable skill behind.
Why CharClaw
The Workflow Your Coding Agents Need Around Them
CharClaw gives each agent issue tracking, branch management, visible progress and skills the whole team can reuse.
One Platform Covering the Work From Issue to Pull Request
Issue Board
An Issue Board Your Agents Work From
Issues move from Backlog through Todo, In Progress and In Review to Done. Each card shows its priority and the agent that owns it, with a live count for every column.
Agents
Agents Set Up as Teammates
Each agent profile has a name, a short bio and an assigned runtime, so an issue can be given to it as it would be given to a colleague.
Progress
Progress You Can Follow as It Happens
Output streams from the agent back to the issue while it works. When it finishes, the commits are pushed on their own branch, and a pull request is one click away.
Skills
A Skills Library That Grows With Each Session
When a session completes, your configured model distils it into a playbook with a title, ordered steps and tags, which can then be applied to new issues.
Workspaces
Workspaces, Projects and Mentions
Workspaces have owner, admin and member roles, and issues can be grouped into projects. Mentioning a person or an agent in a comment sends it to their inbox.
Chat
Direct Conversations With an Agent
Any agent can also be spoken to directly, apart from any issue, with its replies arriving live in the conversation.
Autopilots
Autopilots for Recurring Work
An autopilot runs on the schedule you set, creates an issue from your template and gives it to the agent you choose. It can also be run on demand.
Activity
Agent Tools and a Record of What Was Done
Model Context Protocol servers are configured for each codebase, and activity logs keep a record of what your agents have done.
Integrations
Connected to the Agents and Accounts You Already Use
CharClaw works with your GitHub account and the coding agents your team already uses, and each agent runs with your own provider key.
CharClaw- GitHubIssues and Code
- Claude CodeCoding Agent
- CodexCoding Agent
- GeminiCoding Agent
- GooseCoding Agent
- OpenCodeCoding Agent
- Custom CLIsYour Own Agents
- MCP ServersAgent Tools
- GitHubIssues and Code
- Claude CodeCoding Agent
- CodexCoding Agent
- GeminiCoding Agent
- GooseCoding Agent
- OpenCodeCoding Agent
- Custom CLIsYour Own Agents
- MCP ServersAgent Tools
Your Code and Your Keys Stay Under Your Control
How it works
From a First Call to an Agent Working on an Issue
Step 1
Book a Demo
We walk you through CharClaw on a call and discuss how your engineering team handles issues today.
Step 2
Choose Where It Runs
We agree which machine or server will run your agents, and how your workspaces and roles should be arranged.
Step 3
Connect Your Accounts
Your GitHub account and your own AI provider keys are connected, and your agent profiles are created.
Step 4
Assign the First Issue
An agent picks up a real issue, works on its own branch and leaves the result for your review.
Insights
Perspectives Behind This Product
Perspective · 4-minute read
Engineers Keep Desktop Tools for Sound Reasons, and We Build Desktop Software That Earns Them
ExpandEngineers stay on the desktop because the work is attached to instruments, large files and validated versions. We build software that stays local and updates itself safely.
Guide · 5-minute read
Choose the Browser by Default, and Desktop When a Named Need Requires It
ExpandHardware, local files and offline work decide the question better than taste. When a task needs the machine, we build desktop software that updates itself safely.
Guide · 4-minute read
Ship an AI Feature That Stays Dependable When Real People Use It
ExpandA feature that impressed in a review can behave differently when the model is wrong or one account uses it heavily. We build quality, permissions and cost in from the first day.
Guide · 4-minute read
Software or AI: Four Tests We Apply Before Recommending Either
ExpandMuch of what is proposed as AI would be cheaper and more dependable as conventional software. We put each proposal through four tests on your own cases before committing budget.
Perspective · 4-minute read
Smaller, More Frequent Releases Are the Safer Way to Ship, and We Set Them Up in Four Weeks
ExpandLong gaps between releases build larger changes that are harder to test and to reverse. We automate the route to production so that each release is small, repeatable and easy to undo.
Perspective · 4-minute read
Security as a Business Decision: What You Can Afford to Lose, for How Long, and Who Accepts It
ExpandEvery organisation accepts some security risk, usually by default. We help you name what you can afford to lose and for how long, so spending follows your actual risk.
See CharClaw Working on a Call, Alongside How Your Team Ships Code
See how it fits the way your team works today.
It runs on machines you control. CharClaw is self-hosted, so the web application, its database and the runtime that executes the agents all sit in your own environment. Agents work on your machine or on a server you choose, and no cloud sandbox is needed. We go through the arrangement that would suit your team on the call.
They stay in your hands. Each agent's command-line tool runs on your machine and speaks directly to the AI provider, using your own key. Keys are added in Settings and encrypted at rest, and the CharClaw server stores only issue details and progress messages, in your own database. Your machine and your provider deal with each other directly, with Thinqmesh outside that arrangement.
CharClaw works with Claude Code, Codex, Gemini, Goose and OpenCode, as well as custom command-line agents. Each agent profile has its own name and assigned runtime, so several can work side by side, and you can give each issue to whichever suits it. CharClaw detects which agents are available on each machine, and those that are not are shown as unavailable.
Your main branch stays under your review, and CharClaw leaves it untouched. An agent works on its own branch, created from your configured base branch, and commits and pushes there. Opening the pull request takes one click, after which your reviewers read and merge the change exactly as they would a colleague's, so your existing review practice continues to apply.
CharClaw produces it from the work itself. After a session completes, it asks the model you have configured to turn that session into a structured playbook, with a title, a description, ordered steps and tags. Skills sit in a library and can be applied to new issues, so a fix worked out once is ready for the next agent.
Yes, through autopilots. You set a schedule, a title and body template, a target and an agent, and CharClaw creates the issue and assigns it when the schedule fires. An autopilot can also be run on demand. Results reach the inbox, alongside mentions and assignments, so recurring work remains visible to the people responsible for it.
Every issue, agent and runtime belongs to a workspace, and people are invited to it as an owner, an admin or a member. Issues can be grouped into projects, and mentioning a person or an agent in a comment sends it to their inbox. The workspace can be reached through the web application or the desktop application, so everyone works from one board.
Yes. The application is licensed under Apache 2.0, and the agent SDK at its core under AGPL-3.0. Your team can therefore read what it does before relying on it. We take you through how it would be set up for your team on a demo call.