aisettings
Log InAdd your setup

Dax Raad

Engineer
Last updated: Sep 16, 2026
RoleCreator of OpenCode; SST builder · anomalyco
EarnsEngineer
Sources15 cited fields · medium confidence
@thdxrthdxrthdxr.com

Uses OpenCode with fast Kimi/GLM models for routine work and GPT for large background tasks, while designing the harness for provider switching and parallel orchestration.

“if coding agents become an important tool for building software they can't all be proprietary and vendor locked” x.com

Share cardShare your setupCompare

Loadout

aisettingsShare on X

Dax Raad’s loadout

Engineer
Harness
OpenCode
OpenCode 2. Uses the latest OpenCode 2 model-selection feature to assign a different model to subagents for review.
Model
Kimi
Kimi and GLM for most tasks because they are fast.
Also uses
GPT-5.6 Sol
A portion of the team returned to Sol after evaluating Astra.
ToolsMCPs+4 more

3 views · 0 equipped

Models

Kimi
Kimi and GLM for most tasks because they are fast.
x.com ↗
GPT-5.6 Sol
A portion of the team returned to Sol after evaluating Astra.
x.com, 2026-09 ↗
GPT-6 Astra
Evaluated by the team; Dax says it has novel capabilities but doubles effective spend.
x.com, 2026-09 ↗
GPT
For large tasks he wants to run in the background.
x.com ↗

Tools

MCPs
Adds project-specific MCPs incrementally rather than loading many up front.
x.com, 2026-09 ↗
ElevenLabs + Slack
Uses the ElevenLabs MCP to send funny text-to-speech messages to the team in Slack.
x.com, 2026-09 ↗
OpenCode SDK
Type-safe JS client for launching instances and running tasks programmatically in parallel.
x.com ↗
Provider profiles
Provider-specific tools and system prompts, with future optimized setups per provider.
x.com ↗
Plans and todo lists
OpenCode makes plans and checks them off as it works.
x.com ↗

Workflow

  1. 1Use a different, smarter model for subagent reviews rather than having the primary model review itself.x.com, 2026-09
  2. 2Keep the MCP set small initially, then add project-specific integrations as useful cases appear.x.com, 2026-09
  3. 3Use different models by task: fast Kimi/GLM for most work and GPT for large background jobs.x.com
  4. 4Design OpenCode to be embeddable and composable, then run parallel tasks through the SDK.x.com
  5. 5Keep models interchangeable rather than locking the workflow to one vendor.x.com

Timeline

  1. 2026-09-16HarnessOpenCodex.com, 2026-09
  2. 2026-09-11Also usesGPT-5.6 Solx.com, 2026-09
  3. 2026-09-11Also usesGPT-6 Astrax.com, 2026-09
  4. 2026-09-11ToolsMCPsx.com, 2026-09
  5. 2026-09-11ToolsElevenLabs + Slackx.com, 2026-09

Comments

Log in to comment