Model management

§ MODELS

Octopush is bring-your-own-keys. You connect the providers you already pay for, pick a model with the ModelPicker, and Octopush remembers your choice per workspace. Cost estimates update as you switch, so you always know what the active model costs.

Providers

Octopush wires four providers:

• Anthropic (Claude) — paste your API key.
• OpenAI — paste your API key.
• DeepSeek — paste your API key.
• Ollama (local) — point Octopush at your local Ollama base URL. No API key required.

Configure them in Settings ▸ Integrations.

[screenshot: Settings ▸ Integrations with provider keys and the Ollama base URL]

Note

Those four are the providers Octopush v0.1.70 supports. Anthropic, OpenAI, and DeepSeek use API keys; Ollama uses a local base URL and runs entirely on your machine.

Picking a model

The ModelPicker lets you choose the model for a workspace. Your selection persists per workspace, so each task keeps the model you assigned it. You can also switch mid-session from the picker or via the command palette (⌘K).

As you change models, the cost estimation updates to reflect the new model’s pricing — visible on every message in TALK and in the DIRECT cost ledger.

A typical flow

1. In Settings ▸ Integrations, add at least one provider key (or your Ollama base URL).
2. Open a workspace and use the ModelPicker to assign a model.
3. Work in TALK; watch per-message cost update for that model.
4. Hit a hard step? Switch to a stronger model — the cost estimate adjusts.
5. The workspace remembers your last selection the next time you open it.

Tips

Run free with Ollama

If you want zero cloud cost, point Octopush at a local Ollama instance and select it as your model. Everything — TALK, RUN, REVIEW, and the per-message cost display (which reads as no spend) — works the same.

Note

In DIRECT, each pipeline stage gets its own assigned model, overridable per run. Model management here is about the per-workspace default for TALK; DIRECT layers per-stage choices on top.