microsoft/amplifier-app-opencode

microsoft/amplifier-app-opencode

Description: Amplifier Agent adapter for opencode

Language: Python

License: MIT

Stars: 0

Forks: 0

Open issues: 2

Created: 2026-06-20T22:24:01Z

Pushed: 2026-06-23T03:31:52Z

Default branch: main

Fork: no

Archived: no

README:

amplifier-app-opencode

Launch the opencode TUI on top of amplifier-agent with one command:

amplifier-opencode

This binary discovers what models your local amplifier-agent serves, writes a working opencode config from that discovery, and launches opencode — every time, no manual config to maintain.

---

What this gives you

opencode is a fast terminal coding assistant. amplifier-agent is Microsoft's modular agent framework with a multi-provider OpenAI-compatible HTTP face.

Without this adapter, integrating them requires you to manually mirror amplifier-agent's /v1/models output into your opencode config and re-edit it whenever your provider mix changes.

With this adapter:

• One command (amplifier-opencode) handles the whole flow
• The model list in opencode's picker is always live — re-discovered on every launch
• amplifier-agent's server is auto-started in the background if it isn't running
• Drop-in opencode TUI: /models, /connect, slash commands all work normally
• Includes a doctor subcommand that diagnoses any setup issue before you ask

---

Prerequisites

You need three things installed. The first two are owned by other projects; this README walks through the official install for each so a brand-new machine can get set up start-to-finish.

1. amplifier-agent — the backend server (>= 0.8.0 required)

amplifier-agent is the OpenAI-compatible HTTP server this adapter talks to. Install it via `uv` (Astral's Python package manager):

# Install uv if you don't have it
curl -LsSf https://astral.sh/uv/install.sh | sh

# Install amplifier-agent as a global tool
uv tool install amplifier-agent

# Verify — output MUST report 0.8.0 or higher
amplifier-agent --version

> Version requirement: `amplifier-agent >= 0.8.0` is mandatory. Versions > below 0.8.0 do not ship the serve chat-completions HTTP face, multi-provider > routing, or the auth subcommand — amplifier-opencode will fail to spawn > the server, populate /v1/models, or read persisted credentials. > > If amplifier-agent --version reports an older version, upgrade with > uv tool upgrade amplifier-agent and re-run amplifier-opencode doctor > to confirm.

For full install options (source builds, pinned versions, etc.) see the amplifier-agent README.

2. opencode — the TUI

The official one-line installer downloads the platform-native opencode binary:

curl -fsSL https://opencode.ai/install | bash

It places opencode in ~/.opencode/bin/ and adds that to your shell PATH. Open a new terminal (or source ~/.zshrc) and verify:

opencode --version

For other install methods (Homebrew, manual download, package managers) see opencode.ai/docs/intro.

3. amplifier-app-opencode — this adapter

Same uv-tool pattern as amplifier-agent:

uv tool install --from git+https://github.com/microsoft/amplifier-app-opencode amplifier-app-opencode

# Verify
amplifier-opencode --help

(Once published to PyPI, this becomes uv tool install amplifier-app-opencode.)

4. At least one provider credential

amplifier-agent talks to upstream model APIs (Anthropic, OpenAI, etc.) and needs credentials for at least one of them. Easiest: set an environment variable that amplifier-agent already knows about:

# Pick at least ONE that you have access to:
export ANTHROPIC_API_KEY="sk-ant-..."
export OPENAI_API_KEY="sk-..."
export AZURE_OPENAI_API_KEY="..."
export OLLAMA_HOST="http://localhost:11434" # if running ollama locally

# To persist across all terminals, add the line to ~/.zshrc (or your shell's rc)

Alternative: use amplifier-agent's persistent credential file. Run once and your key is stored under ~/.amplifier-agent/credentials.json (mode 0600), available to every future invocation from any directory:

amplifier-agent auth set anthropic sk-ant-...
amplifier-agent auth list # confirm it's stored

The amplifier-agent server uses env-first resolution: shell env vars win over the credentials file, so existing shell-rc workflows continue working unchanged.

---

Provider Credentials

amplifier-opencode auto-detects which AI provider you have credentials for and configures amplifier-agent to serve those providers. Set one or more of the following environment variables before running amplifier-opencode launch:

| Provider | Env var | |---|---| | Anthropic (Claude) | ANTHROPIC_API_KEY | | OpenAI (GPT) | OPENAI_API_KEY | | Azure OpenAI | AZURE_OPENAI_API_KEY or AZURE_OPENAI_KEY | | Ollama (local models) | OLLAMA_HOST |

Run amplifier-opencode doctor to see which providers will be available.

Advanced: custom host_config.json

If you want fine-grained control (custom MCP servers, approval policies, per-provider config overrides), write your own host_config.json and pass it with --host-config:

amplifier-opencode launch --host-config /path/to/host_config.json

See amplifier-agent's host_config documentation for the full schema.

---

First run

amplifier-opencode

You'll see something like:

[1/4] Starting amplifier-agent (port 9099, workspace='opencode')
amplifier-agent ready at http://127.0.0.1:9099/v1
[2/4] Discovering models via GET http://127.0.0.1:9099/v1/models
- anthropic claude-haiku-4-5-20251001 Claude Haiku 4.5
- anthropic claude-opus-4-8 Claude Opus 4.8
- anthropic claude-sonnet-4-6 Claude Sonnet 4.6
[3/4] Wrote /Users/you/.config/opencode/opencode.jsonc (global config)
[4/4] Launching opencode in /Users/you/...

The opencode TUI then opens. Type /models and pick a Claude model under the Amplifier section. Send a prompt. Done.

---

Daily usage

Just run it

amplifier-opencode

The default action does the full check → discover → write → launch flow. From any directory.

Pass arguments through to opencode

Anything after -- is forwarded to opencode:

amplifier-opencode launch -- run "summarise this...