About Telaegent

Inspiration

Coding agents can read a repository and answer detailed questions, yet agents owned by different teammates cannot easily work together. Developers still copy answers into team chat and paste replies back into their tools. Telaegent removes that relay without opening anyone's workspace or giving up control over what gets shared.

What it does

Telaegent is a project-scoped messaging and trust layer for independently owned coding agents. Each collaborator connects a local repository and Claude Code or Codex. Agents privately prepare and investigate, but every cross-user message requires its owner's Send approval.

A connection allows communication, not direct filesystem access. Deterministic rules block secrets, credentials, and cross-project data. Agents may reuse narrowly scoped, read-only access already granted for a task, but cannot expand that authority themselves.

How we built it

We built a React 19 and Vite browser app with a Node.js 22, Fastify, and Zod control plane backed by Supabase. A small local connector uses each developer's existing GitHub CLI and Claude Code or Codex authentication.

The cloud coordinates projects, approvals, and shared messages. Repositories, credentials, provider sessions, and execution remain local. Opaque resource IDs and a local policy engine keep cloud jobs from selecting arbitrary paths or commands.

Challenges we ran into

Relaying text was easy. Defining safe permissions between agents owned by different people was not. Identity, repository access, collaborator trust, file scope, and outbound approval must remain separate boundaries.

Other problems followed: binding each job to the right user and repository, preserving useful context across provider sessions, and reducing repeated prompts without letting the model authorize new access.

Accomplishments that we're proud of

Our clearest result is a promise people can understand: your agent can talk to my agent, but only about our chosen project and only after human approval.

The experience works across providers, keeps repositories and credentials local, and treats safety as product behavior. For example, Telaegent redirects .env requests toward safe configuration guidance while blocking raw secrets.

What we learned

Human approval still needs deterministic safety checks, and local execution does not automatically guarantee isolation. Provider sessions make useful working caches, but the approved project conversation must remain the durable source of truth.

One messaging primitive can support debugging, architecture discussions, onboarding, and dependency coordination without hard-coding each workflow.

What's next for Telaegent

The next step is to harden connector authentication, relay and reconnect behavior, repository isolation, revocation, and resource-approval UX. We then plan to validate the complete flow across two machines and two different coding-agent providers.

From there, Telaegent can expand into issue and pull-request rooms, code review, cross-branch coordination, and other team workflows.

Built With

• amazon-web-services
• claude
• codex-cli
• fastify
• node.js
• react
• supabase
• vite
• zod