Capture the contrast
Keep the change in approach, not just the error. Woodwide looks for corrections in supported sessions so you can revisit what changed. It may miss a correction or important context.
Free beta · Unlimited corrections & checks
Keep what your
coding agent learned.
Woodwide is correction memory for AI coding agents. Keep readable playbooks from completed coding sessions and corrections, then recall relevant lessons for your next task.
No Woodwide account needed for local setup.
Extraction uses your configured cloud model provider and may incur charges.
Readable JSON + MarkdownAccount-free local modeOptional reviewed sharing
From correction to context
Keep the lesson.
Not just the transcript.
A playbook records the problem, attempted steps, pitfalls and reported outcome. Conditions and verification help you decide whether it applies again.
Keep a readable copy
A screened summary is saved on your machine as JSON and Markdown. Read, search and export existing memory without another model call or server connection.
Recall without a pause
Private recall looks for relevant history first. In manual mode, a playbook offer never pauses agent work. Unanswered offers expire after 60 seconds unless you interact with that suggestion before it expires.
Illustrative lesson · not a live result
More useful than
“try that again.”
A short observation is useful when it includes the boundary of the advice. It is not permission to run a command blindly.
example / install failureIllustration
Check the required runtime before retrying.
- When
- An install fails because the active runtime does not meet the project’s declared requirement.
- Instead
- Select the project’s required runtime, then retry the original install.
- Verify
- Confirm the active version and rerun the failing command.
- Except
- Do not assume this explains unrelated network or permission failures.
Choose the boundary
Local memory first.
Sharing is a separate choice.
Private local memory
Keep and search personal summaries on your machine. Local setup switches off sharing and benchmark telemetry.
Local storage is not on-device extraction. A redacted digest goes to your chosen cloud provider when extraction is enabled.
Set up local memory →Reviewed shared lessons
Review generalized contributions before publishing. Network recall sends a task or error query and stack context to the Woodwide server.
Sharing, extraction and telemetry are separate controls. Automated redaction reduces risk; it cannot guarantee confidentiality.
Continue with Google ↗Woodwide supplements transcripts and project instructions; it does not replace your agent or independently verify a fix. Read the data boundaries →
A few things worth knowing
Before you begin.
What is free in the beta?
Woodwide beta use includes unlimited corrections and checks. Your extraction provider’s subscription limits, API charges and terms still apply. Unlimited Woodwide use does not mean free or unlimited model inference.
How is this different from keeping transcripts?
Transcripts retain the session history. Woodwide derives a shorter playbook with the problem, steps, pitfalls, lessons and reported verification, then provides local search and recall. The shorter record can miss context; inspect the evidence before reuse.
Do I need to sign up or share my work?
No Woodwide account or server is needed for private local setup. Cloud extraction is separate from sharing. Google sign-up is optional and available only where linked; an account lets you connect to the network without requiring contributions. You do not need it to read, search or export existing local summaries.
Will suggestions interrupt my agent?
No pause for a decision. In the default manual mode, Woodwide gives one optional playbook suggestion and your agent keeps working. An unanswered offer expires after 60 seconds unless you explicitly interact with that suggestion before it expires. Unrelated chat or tool activity does not extend it; silence is not acceptance.
Can my agent use guidance automatically?
Auto-consume is off by default. Enable it to accept eligible, screened guidance in the current task without another Woodwide confirmation. It does not launch a separate agent or grant permission to use tools, make destructive changes or upload data. See the enable and disable commands.
Which coding agents are supported?
Claude Code has hooks and a bundled plugin; use one integration, not both. Codex hooks need your review in its /hooks interface. OpenCode has a plugin. Gemini CLI supports transcript backfill and MCP, rather than live hooks. Choose and review your agent integration.
Is all of Woodwide open source?
The core, CLI and Claude Code plugin are Apache-2.0. The server is source-available and proprietary, under a separate license.
A smaller next step
Start with something
you have already learned.
Get started with Woodwide Private setup first. A model call only when you enable extraction.