From idea to pull request, in minutes.

Open source coding orchestration for teams shipping production-ready code with AI.

Introducing Ivy Tendril

Ivy Tendril is the ultimate agentic workflow tool. Optimize workflows across multiple agents, tasks, codebases, and GitHub repositories. Point it at issues and watch it plan, implement, test, and open a PR, all while continuously improving over time.

Works with

From draft to ship

Each step in the lifecycle has its own surface — drafting a plan, watching it implement, reviewing the report.

Create a plan

Start with a prompt. The plan captures intent and constraints before any agent picks it up.

Apps: Drafts

No active plan

Approve to implement

An implementation agent picks up the plan, streaming tool calls and reasoning as it works.

Apps: Jobs

#142

Ready

Speed up cold-start latency

Review the report

Switch between verifications — tests, lint, diff — to see exactly what changed and what's green.

Apps: Review

Review report — Speed up cold-start latency

Tests

Lint

Diff

✓ auth.spec.ts (14 tests)✓ jwks-cache.spec.ts (6 tests)✓ middleware.spec.ts (8 tests) 28 passed · 0 failed · 2.1s

Agents are promptware

Each agent is shaped by a program, a set of tools, and memory it can grow as it works.

Program, tools, memory

An agent is defined by what it knows how to do, what it can reach for, and what it remembers.

Concepts: Promptware

implementer

program.md

tools/

memory/

Writes back to itself

Agents revise their own program and append to memory as they learn from each plan they ship.

Concepts: Memory

program.md

Implement plans by editing files.

memory/notes.md

· JWKS responses cache well for 10 minutes.

Reads before it runs

Before executing a plan, the agent loads its program and consults memory for relevant context.

Concepts: Execution

PlanSpeed up cold-start latency

implementer reads context…

program.md

memory/notes.md

Plans evolve with you

Refine the goal, split scope when it grows, and stream every job as agents work.

Refine in place

Update the plan and Tendril rewrites the body to match — so intent stays fresh as scope shifts.

Concepts: Plans

#142

Draft

Speed up cold-start latency

Profile the auth handler and see if anything stands out.

Split when scope grows

One plan becomes two when the work is bigger than expected — dependencies wired automatically.

Concepts: Plans

#142

Draft

Speed up cold-start latency

Stream every job

Watch planning and implementation jobs side by side. Open any output to follow the agent live.

Apps: Jobs

Jobs

planning

Refine cold-start latency planOutput

implement

Cache JWKS responses

Output

implement

Add cold-start benchmarkOutput

How does it work?

Ivy Tendril is built on Promptware, self-improving software agents with specific tasks, knowledge, tools, and memory. By designing each part of the development cycle with dedicated Promptware, your workflows learn, adapt, and get better with every run.

Load

Agent reads its evolving Program.md — the living source code

Learn more about Promptware

Get Started

Download the software, connect it to one or more GitHub repositories, and start working. Ivy Tendril works with Claude Code, Cursor, Windsurf, and more.

macOS / Linux

~curl -sSf https://cdn.ivy.app/install-tendril.sh | sh

Windows

~Invoke-RestMethod -Uri https://cdn.ivy.app/install-tendril.ps1 | Invoke-Expression

.NET Tool

~dotnet tool install --g Ivy.Tendril

Then run:

~tendril

Read the full docs

Get Involved

Tendril amplifies great developers, it doesn't replace them. Contribute, join the conversation, or stay in the loop.

GitHub

Star, fork, and contribute to the framework

Discord

Chat with the community and get help