Several reviewers. One review on the thread.
Agents you name — Legal, Compliance, Team Lead, Senior Engineer, or your own — fan out on the same diff. Findings merge so two agents flagging the same line become one comment. Each still posts its own check.
How it works
Connect the repository. The board reviews the diff. You keep the merge.
PR #482
ConnectedKeys stay yours. TeraCodeAI never resells inference.
1.
Connect
Install the GitHub App, or connect a GitLab project with a token and a webhook. Add an Anthropic or OpenRouter key to the vault — TeraCodeAI never resells inference.
2.
Review
The agents you name fan out on the same diff, in parallel, each with its own instructions, model and skills.
3.
Ship
Findings merge, so two agents flagging one line become one comment. The review posts on the pull request and each agent posts its own check.
PR #482
ReviewingMerged from 4 agents
Agents:
A board, not one generalist.
Each agent has its own instructions, model, and skills. They run in parallel. Disagreements are visible; duplicates are not posted twice.
Ranked by risk
Ranked by what it would break.
A logged credential and a missing trailing comma are not the same finding. Severity is something you can act on, not a line-number sort.
One thread, several checks.
Comments land as one review from the App. Checks stay per agent, so you can require security without requiring style.
Posted to the pull request
ReviewedNamed reviewers on the diff, merged into one thread.
Attach instructions to file globs so the agent reading auth is not the one reading a migration.
The host calls your provider. Tokens are billed to you. We add nothing on top.
Quality is which findings the team resolved rather than deleted — not how confident the model sounded.
Skills attach to paths
auth/session.ts
auth · tokens
api/orders.ts
hot path
db/migrate.sql
schema change
terms/dpa.md
policy
What it looks for
Agents you name. One review on the thread.
Security
Secrets in logs, broken auth, missing rollback paths — the findings that actually ship incidents.
Legal and Compliance
The clauses and controls your team has to answer for, read on the diff rather than at audit time.
Team Lead
House rules and the shape of the change: is this the way this codebase does it?
Or your own
An agent is instructions, a model and skills. Name one after whatever your reviews keep repeating.
Who it is for
Built for teams that already generate more code than they can read.
AI-authored pull requests
Agents write faster than any review rotation. This is the board that lets that speed reach main.
No seat tax
One meter: the first repository in an organization is free forever, then $20 per connected repository per month. There is no per-developer licence.
Security as a reviewer, not a linter
A logged credential and a missing comma are not the same finding. Risk order is the product.
Works with what you already have
Your forge. Your provider. Your keys.
TeraCodeAI connects to the source control you already use and calls the model you already pay for. We do not list forges we have not built.
GitHub
Source control · GitHub App
GitLab
Source control · Token + webhook
Anthropic
Model provider · Your API key
OpenRouter
Model provider · Your API key
Security and keys
You keep the keys. We never sit between you and the model.
Bring your own key
Reviews call Anthropic or OpenRouter with the key you put in the vault. Tokens land on the invoice you already reconcile.
No hosted inference
There is no trial model behind the curtain. Without a key there is no review — only a labelled fixture of what a comment looks like.
Scoped to the diff
The board reads the change you asked it to review. It does not browse the rest of your estate.
The review is the only write
Comments land as one review, plus a status check per agent. Nothing auto-merges and nothing is pushed to your default branch.
Why not the usual reviewer
Most AI reviewers bill per seat and resell you tokens.
| Typical AI reviewer | TeraCodeAI | |
|---|---|---|
| Who pays for inference | They resell you tokens | Your provider, at their price |
| Platform fee | Per seat, per month | $0 first repo, then $20 each |
| Who reviews | One generalist | A board of agents you name |
| What lands on the PR | A comment storm | One review, one check per agent |
| How findings are ordered | By line number | By what it would break |
| Duplicate findings | Posted twice | Merged into one comment |
| What it writes | Comments, sometimes commits | The review. Nothing else. |
Every AI reviewer charges per seat. TeraCode runs on your keys, at cost.
$0
Markup on your token spend
0
Per-seat licences
100%
BYOK — your own API keys
1
Review on the pull request
Questions about Review
Start free connects your first repository. It stays free, forever.