Which providers and models does Terse track with opencode?

opencode is provider-agnostic, and so is Terse's tracking. Whether you point opencode at Claude Sonnet/Opus, GPT-4o, Gemini 2.5, a model through OpenRouter, or a local model via Ollama, Terse reads the active model and shows per-turn cost at that model's real input/output rate. Switch providers mid-project and the cost tracking follows.

Reduce opencode Token Costs — Terse for opencode

Q: Does Terse work with opencode?

Yes. Terse auto-detects the opencode process via process scanning and attaches a floating monitor to your terminal. It compresses the prompts you type before they reach whatever model opencode is calling, flags redundant file reads as opencode's agentic loop re-reads the repo, and tracks per-turn token cost. Because opencode is provider-agnostic — Claude, GPT, Gemini, OpenRouter, or a local model — Terse reads the active model and applies the correct pricing.

Q: Why is my opencode session so expensive?

opencode is agentic: each turn it reads files, runs commands, and applies diffs, then feeds all of that output back into context for the next turn. That context accumulates every turn, so a long session can push tens of thousands of tokens per request. Because you bring your own API key, that growth is billed directly to you. Terse compresses your typed prompts and flags when opencode re-reads the same file so you can interrupt before the context — and the bill — snowballs.

How Terse Works with opencode

opencode runs an agentic loop: it reads your repo, edits files, and runs commands, feeding every diff and command output back into context each turn. That context — and the bill on your own API key — grows fast. Terse compresses what it can (your typed prompts) and flags what you shouldn't ignore: the same files being re-read turn after turn.

Terminal — opencode Session

Terse Active

$ opencode

opencode · provider: anthropic · model: claude-sonnet-4-5 · repo indexed

● Terse detected opencode — monitoring (Sonnet 4.5 · $3.00/M input)

> could you please go ahead and update the payment webhook handler so that it retries on failure and also i think we should probably add some logging around it just to be safe

opencode reading: src/payments/webhook.ts [2,480 tokens]

⚠ Terse: payments/webhook.ts read 3× this session — it's already in context, re-read is wasted tokens

12.6Kprompt tokens

34.1Kwithout Terse

63%compressed

$0.05this turn

⚡

Auto opencode Detection

Terse scans running processes every 5 seconds. When opencode launches in your terminal, a floating monitor bar appears automatically — no plugin, no config, no changes to opencode itself.

Automatic

🗜️

Prompt Compression

Every instruction you type into opencode's prompt goes through Terse's 7-stage pipeline: typo correction, filler removal, hedge stripping, and phrase shortening. Code blocks and file paths pass through unchanged.

7-Stage Pipeline

📁

Redundant File Read Flagging

opencode's agentic loop re-reads files as it edits — often the same file several times per session. Terse tracks this and warns when a file has already been read, so you can stop it re-loading context it already has.

Session Tracking

🔀

Provider-Agnostic Cost Tracking

opencode works with Claude, GPT, Gemini, OpenRouter, or a local model. Terse reads whichever model is active and tracks per-turn cost at that provider's real rate. Switch providers mid-project and the numbers follow.

Per-Turn Visibility

📊

Session Cost Summary

At the end of each opencode session, Terse shows total token usage, total cost on your own API key, and a breakdown of where tokens went — prompts, file reads, command output. Know exactly what drove the bill.

Session Analytics

🔒

100% On-Device

Terse reads your terminal input via OS Accessibility APIs and compresses locally in Rust. No prompt text, no code, and no file contents are ever sent to Terse's servers — fitting for an open-source, self-hosted workflow.

Zero Cloud

Supported opencode Providers

opencode is provider-agnostic, and so is Terse. It tracks token costs and monitors sessions regardless of which model you point opencode at — directly or through OpenRouter.

Claude Sonnet 4.5

$3/M input · $15/M output

Claude Opus 4

$15/M input · $75/M output

GPT-4o

$2.50/M input · $10/M output

Gemini 2.5 Pro

$1.25/M input · $10/M output

GPT-4o mini

$0.15/M input · $0.60/M output

Gemini 2.5 Flash

$0.075/M input · $0.30/M output

Claude Haiku 4.5

$0.80/M input · $4/M output

Local (Ollama)

Token count monitoring · no cost tracking

Frequently Asked Questions

Everything about using Terse with opencode.

Does Terse work with opencode?

Yes. Terse scans running processes every 5 seconds. When opencode launches in your terminal, Terse detects it and activates a floating monitor bar — no plugin and no changes to opencode. It reads the active provider and model, so cost tracking works whether you're on Claude, GPT, Gemini, or a local model.

Why is my opencode session so expensive?

opencode is agentic — each turn it reads files, runs commands, and applies diffs, then feeds that output back into context for the next turn. That context accumulates every turn, and since you supply your own API key, the growth bills straight to you. Terse compresses your typed prompts and flags redundant file reads so the context doesn't snowball unnoticed.

Which providers and models does Terse track?

All of them that opencode supports. Point opencode at Claude Sonnet/Opus, GPT-4o, Gemini 2.5, an OpenRouter model, or a local model via Ollama, and Terse reads the active model and shows per-turn cost at that model's real input/output rate. Change providers mid-project and the tracking follows automatically.

Does Terse interfere with opencode's agentic editing?

No. Terse only compresses the prompts you type into the terminal input. It never touches opencode's file edits, the commands it runs, the diffs it applies, or the context it builds. Your agentic loop runs exactly as before — you just send shorter, cleaner instructions into it.

Cut Your opencode Token Costs by 40–70%