Repository navigation
feature: OpenCode support — detection, settings, parsed events, token and cost accounting #941
Description
Feature Description
Second agent on the agent integration contract, and the first non-Claude one. OpenCode is
open source, so it is the integration anyone outside the team can reproduce.
Proposed Solution
- Detect an installed OpenCode and locate its settings
- Patch it to use
localhost:47600and the Cortex CA; ask before editing, support revert - Traffic parsed into typed events, not merely proxied
- Token counts and cost correct for its wire format; if it is not Anthropic-shaped, either
parse it properly or emit a clearly-labelled partial event - Session and cost metrics aggregate alongside Claude Code sessions, with field names
agreed with per-session token and cost aggregation - A tool inventory for OpenCode, or state explicitly that tool pruning is unsupported for
it — without one, pruning silently saves nothing
Required artifact: a doc for OpenCode
This issue does not close on code alone. The doc is a deliverable like the code: it ships
in the same PR, or in a linked PR merged before this issue closes. A PR description or a
comment here does not count.
It is docs/agents/opencode.md in this repo, linked from the root README's Any agent works line
and from the index in docs/agents/README.md (#939). It covers:
- Enable and revert — the
agentop configurecommands for OpenCode, which file each one
writes, and how to undo it by hand - CA trust — what OpenCode needs to trust the Cortex CA, and the symptom when it does not
- What Cortex shows for it — which events are typed, which token and cost fields are
filled, and what is partial or unsupported (e.g. tool pruning) - Verified depth — the OpenCode versions and platforms actually tested, and what was not
- Known issues specific to OpenCode
cmd/agentop/README.md can carry the command reference too; it does not replace this page.
Additional Context
Part of the Cortex v0.9.0 release. Depends on the agent integration contract.
Activity
esnible commentedon Sep 11, 2026
See https://opencode.ai/ for OpenCode install instructions.
Those instructions mention ~/.config/opencode/opencode.json perhaps it is ~/.config/opencode/opencode.jsonc ?
The content in that file became
{ "$schema": "https://opencode.ai/config.json", "provider": { "litellm": { "npm": "@ai-sdk/openai-compatible", "name": "LiteLLM", "options": { "baseURL": "https://ete-litellm.ai-models.vpc-int.res.ibm.com" }, "models": { "gpt-5.6-terra": { "name": "GPT-5.6 Terra" }, "claude-sonnet-5": { "name": "Claude Sonnet 5" }, "deepseek-chat": { "name": "DeepSeek Chat" } } } } }I then started OpenCode with
abctl exec -- opencodeI was now able to
/connectwithlitellmand supply my LiteLLM API key interactively.I then was able to interact and had no trouble observing token counts in abctl.
Alan-Cha commentedon Oct 2, 2026
Noting #1239 (OpenAI Responses API support in
inference-parser) as a related issue — surfaced while scoping #942 (Codex), but relevant here too since it's about a wire-format gap in the shared parser, not an agent-specific one.feat: Recognise OpenCode's User-Agent and read its session header
feat: Add peerproc.Environ, the environment a process started with
fix: Report a withheld or unreadable environment as ErrNotFound
4 remaining items
fix: Leave a value disable did not set alone, whatever the record says
fix: Recognise Cortex's proxy by listener, not spelling, in configure…
docs: Add the OpenCode agent page
fix: Say that configure opencode stops OpenCode's service, and ask first
fix: Group OpenCode by X-Session-Id, as the laptop installer already …
fix: Restart OpenCode's service once after configure opencode changes it
docs: Say how OpenCode is grouped and how configure opencode restarts it
docs: Record the live run of configure opencode's restart
fix: Read X-Session-Id by default on laptop installs only
fix: Say only what configure opencode knows; keep the user's own prox…
fix: Name the restart a failed restart needs; say why X-Session-Id is…
fix: Parse OpenCode Zen's Anthropic-dialect endpoints
fix: Price OpenCode Zen's free models at zero
huang195 commentedon Oct 5, 2026
Closing: every criterion is met on
main(d1ca6102). #1243 said "part of #941", so this never auto-closed.agentop configure opencode enable|disable|statusasks before editing (--yesskips the prompt), and disable removes only Cortex's values; the prior state is kept in~/.cortex/opencode-state.json./v1/messagesand/zen/go(Fix: Parse and price OpenCode Zen's endpoints #1247), and any path prefix (Fix: Parse known inference dialects under any path prefix #1248).pricing:entry is added.opencodeagent row, grouped byX-Session-Id(Feat: Support OpenCode as a coding agent #1243). Header-less traffic is attributed by process (Feat: File header-less traffic by the process that sent it #1230).docs/agents/opencode.mdcovers enable/revert, CA trust, what Cortex shows, verified depth and known issues. It is linked from the README's Any agent works line.Not covered here:
docs/agents/README.mdis part of feature: agent integration contract — reusable detection, settings patching, and revert #939, which creates that index./zen/v1/responses(GPT and Grok models) is recorded but not parsed yet; that is feature: OpenAI Responses API support in Cortex's inference-parser #1239.