Cody

You will like Cody if your team runs Sourcegraph, because its context is code search across remote repositories, so chat knows about the service three teams over that nobody on your team has opened. That is the daily trait: answers that cite files outside your checkout. Nobody else can buy it: the Free and Pro plans are gone and the only entry point is an Enterprise contract.

Pick it if Sourcegraph is on the invoice, and use it as the question-answering layer next to a real editing agent. Pick Augment otherwise; it is the closest thing to the same index you can buy alone.

The risk is capability. What is sold today is chat with code-search context and single-file edits: no multi-file edit, no git operations, no terminal execution. Buying that as an agent means the hard part of the task, the change that spans six files and a migration, is still yours, and the tool's job ends where the work begins. An assistant priced as an agent is a category error on the invoice.

Evaluate it as search with a chat box, and it scores well. What it does right: the context comes from a real code-search index rather than a vector guess, and that shows on cross-repository questions.

No benchmark is claimed. Context is retrieved from Sourcegraph's code search across local and remote repositories, a deterministic index rather than an approximate one, which means the same question returns the same files and retrieval errors are reproducible and therefore fixable. What is absent is the loop: multi-file editing and git operations are not offered, so there is no apply-verify-iterate cycle to assess, and retrieval quality cannot be measured through outcomes.

The consequence is that Cody can only be evaluated as a retrieval system. The observation: the best retrieval on this board is attached to the least agent.

Killing free and prosumer tiers is not confidence; it is focus. Sourcegraph ended Cody Free and Pro in July 2025 and archived the public repository as a snapshot, which says the agent is no longer a product, it is a feature, and features do not get their own pricing pages. The moat is the code-search index, which took years to build and is the only thing here a competitor cannot copy in a quarter.

Likely acquirer: a DevOps or observability vendor that wants the graph and will keep the chat as a demo. Position: fine if already in the contract; do not enter the contract for it.

The demo is chat that knows about the other repositories, which is genuinely useful in a large org where nobody has read everything. Procurement is simple: Cody comes only with Sourcegraph Enterprise, from a $16K minimum annual contract, so seats, retention and deployment are whatever that contract already says, and the security questionnaire was answered when code search was bought. Onboarding is an extension install for anyone already logged in.

Approved with conditions: only inside an existing Sourcegraph agreement, never as a standalone purchase, because $16K for an assistant that does not edit is not a purchase I would defend.

cody-public-snapshot is an archived copy of the source, frozen when Sourcegraph went Enterprise-only. I can read it; I cannot run it against anything without a contract, because the useful half is the index on their side. Inside a contract I could bend cody.mcpServers with a disabledTools list, bring my own key, and point it at local models, which is a surprising amount of freedom for a product sold this way.

A fork of the snapshot survives Sourcegraph the way a fossil survives the animal: proof of shape, nothing that moves. Grudging respect for the local-model support; none for the wall in front of it.