You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Add an Integrations page that generates oMLX-style one-shot launch commands for supported CLI agents.
The first version should focus on commands shaped like:
For llama-swap, the executable path should resolve to the appropriate packaged or local CLI path, but
the user experience should stay the same: select tool, model, API key, copy command, run agent.
User Story
As a local model user,
I want llama-swap to generate ready-to-run CLI agent launch commands,
so that I can start Codex, OpenCode, or Pi against my configured llama-swap models without hand-
writing tool-specific config.
Acceptance Criteria
Given the user opens Integrations, when models are configured, then they can choose a model or
alias.
Given the user selects Codex, OpenCode, or Pi, then the UI generates a launch command using
launch --model --api-key .
Given model IDs, executable paths, or API keys contain spaces or punctuation, then generated
shell commands quote arguments safely.
Given no API key is entered locally in the browser, then the command shows a placeholder rather
than exposing stored secrets.
Given OpenAI and Anthropic integrations diverge later, then base URL rules remain explicit:
OpenAI defaults to /v1, Anthropic defaults to server root when the client appends /v1/messages.
Generated integration metadata is testable and does not call external services.
Developer Tasks
#
Task
Effort
Files / Modules
1
Define integration registry schema for launch commands, support level, API family, snippets, and base URL rules
M
proxy/, config/
2
Add initial registry entries for codex, opencode, and pi
S
proxy/
3
Add /api/integrations endpoint returning registry, models, aliases, base URLs, and resolved CLI executable
M
proxy/
4
Implement safe shell-quoting for executable path, model, and API key placeholders
S
proxy/
5
Add Integrations UI page with tool selector, model selector, API key input, command preview, and copy button
M
ui-svelte/
6
Add focused tests for registry validation, command rendering, and shell quoting
Summary
Add an Integrations page that generates oMLX-style one-shot launch commands for supported CLI agents.
The first version should focus on commands shaped like:
For llama-swap, the executable path should resolve to the appropriate packaged or local CLI path, but
the user experience should stay the same: select tool, model, API key, copy command, run agent.
User Story
As a local model user,
I want llama-swap to generate ready-to-run CLI agent launch commands,
so that I can start Codex, OpenCode, or Pi against my configured llama-swap models without hand-
writing tool-specific config.
Acceptance Criteria
alias.
launch --model --api-key .
shell commands quote arguments safely.
than exposing stored secrets.
OpenAI defaults to /v1, Anthropic defaults to server root when the client appends /v1/messages.
Developer Tasks
PRD.md
SPEC.md