Question 1

What is Tendril?

Accepted Answer

Tendril is a desktop app that makes ad-hoc AI coding structured, verifiable, and self-improving. It manages AI coding agents with a plan-based lifecycle, two human checkpoints, automated verifications, and continuous agent learning. It brings structure to AI development like CI/CD did for deployment.

Question 2

Who is Tendril for?

Accepted Answer

Tendril is for any size development team adopting structured AI coding. It's ideal for teams that need local-first, multi-stack support or want to avoid vendor lock-in. Tendril scales from solo developers to large engineering teams.

Question 3

What problem does Tendril solve?

Accepted Answer

Raw AI coding agents are inconsistent and lack oversight. Tendril turns them into orchestrated workers with persistent memory, automated quality checks, and targeted human control.

Question 4

What does the workflow look like?

Accepted Answer

```
Idea/Ticket → CreatePlan → Draft → Human Review → Expand/Split/Update → Execute → Verify → Human Review → PR → Merge
```

There are exactly **two human touchpoints**:

1. Review the plan before execution
2. Review the diff before PR creation

Everything else is automated.

Question 5

What are promptwares?

Accepted Answer

A promptware unit is a self-contained, self-improving agent. Each one contains:

- **Program.md** — Task-specific instructions that evolve over time
- **Memory/** — Persistent learnings across sessions
- **Tools/** — Scoped permissions defining what the agent can access
- **Logs/** — Full execution history for auditability

Tendril ships with built-in promptwares for each stage of the lifecycle: CreatePlan, ExpandPlan, ExecutePlan, UpdatePlan, SplitPlan, CreatePr, and CreateIssue.

Question 6

Which AI coding agents does Tendril support?

Accepted Answer

Tendril is agent-agnostic. Currently supported agents include:

- Claude Code (Anthropic)
- OpenAI Codex CLI
- Gemini CLI

On customer request, we can make Tendril work with any CLI-based agent.

Question 7

Which AI models can I use?

Accepted Answer

Tendril is model-agnostic. Supported models include:

- Claude Opus, Sonnet, Haiku (Anthropic) - through Claude Code.
- GPT-4o, o3 (OpenAI) - through Codex.
- Gemini 2.5 Pro (Google) - through Gemini.

On customer request, we can make Tendril work with basically any model.

Question 8

Can I use my own API keys?

Accepted Answer

Yes. Tendril runs locally and uses your own API keys for whichever LLM provider you choose. There is no Tendril-hosted proxy or middleman. You maintain full control over your API relationship, usage limits, and spend.

Question 9

Can I switch agents or models without disrupting my workflow?

Accepted Answer

Yes. Plans, memory, and verification gates are independent of the agent and model. Switch providers at any time — your accumulated knowledge and workflow configuration carry over completely.

Question 10

What tech stacks does Tendril support?

Accepted Answer

Tendril is stack-agnostic. One Tendril instance manages all your projects regardless of language or framework.

Question 11

Is Tendril cloud-based?

Accepted Answer

No. Tendril is a local-first desktop application. Your code never leaves your machine. The only external calls are to the LLM API (which you choose) and GitHub. This makes it ideal for regulated industries and teams with strict data sovereignty requirements.

Question 12

What platforms does Tendril run on?

Accepted Answer

Windows, macOS, and Linux.

Question 13

How is Tendril different from Copilot, Cursor, or Windsurf?

Accepted Answer

Those tools operate inside your IDE and assist with inline code generation (autocomplete, chat, inline edits). Tendril operates above that layer — it manages the full lifecycle from ticket to merged PR, with plans, verification gates, and agent memory. You can use Tendril alongside any IDE tool. They complement each other.

Question 14

How is Tendril different from Devin or Factory AI?

Accepted Answer

Devin and Factory are cloud-hosted autonomous agents. Your code goes to their servers, you're locked to their models, and there's limited human oversight in the loop. Tendril is local-first (code stays on your machine), agent-agnostic (you pick the agent and model), and built around two defined human checkpoints rather than full autonomy.

Question 15

What does Tendril cost?

Accepted Answer

Tendril is open-source and free.

We offer enterprise companies help with setup and support plans.

Question 16

Are there usage limits or credits?

Accepted Answer

No. Tendril itself has no usage limits, token caps, or credit system. You use it as much as you want. The only variable cost is your LLM API usage, which you pay directly to your chosen provider at their rates.

Question 17

Does my code leave my machine?

Accepted Answer

No. Tendril is a desktop application with a local-first architecture. The only external communication is with the LLM API of your choice and GitHub. Your code, plans, and agent memory all stay on your machine.

Question 18

Do you train on my code?

Accepted Answer

No. Tendril never sees your code. It runs locally on your machine and orchestrates agents that also run locally. The only external calls are to the LLM API you choose and GitHub. Whether the LLM provider trains on your code depends on their terms and your API agreement — Tendril itself stores nothing externally.

Question 19

How long does it take to set up?

Accepted Answer

Ivy offers a 3-day on-site implementation engagement:

| Day | Focus |
|---|---|
| **Day 1** | Foundation — install, configure, connect ticket systems, set up verification pipelines |
| **Day 2** | Training — onboard team, define quality criteria, implement self-improvement loop |
| **Day 3** | Handoff — stabilize, culture workshop, document, transfer ownership |

For self-service, you can get started immediately.

Frequently Asked Questions