Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
36 changes: 36 additions & 0 deletions CHANGELOG.md
Original file line number Diff line number Diff line change
Expand Up @@ -5,6 +5,42 @@ All notable changes to this project will be documented in this file.
The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).

## [0.12.11] - 2026-11-16

**You can use a Claude Pro or Max plan through your own Claude Code login, and workspace review works from every client.** The new Claude subscription provider runs Claude models through the `claude` you already have installed; Altimate Code never reads or stores its token. The first-run picker and `/connect` suggest the subscription logins already on the machine. On workspaces with review on, a member's skill goes to the owner or an admin for approval, everyone sees their own requests and their status, and requests can be withdrawn or removed from the TUI, `serve` and the plugins. Team rules also reach Claude Code and Codex sessions through the plugins. `impact_analysis` answers where a column's value goes, or where it comes from, hop by hop. Also in this release: a capped agent ends with an answer, one overflow compacts once, and the dbt adapter starts under Bun.

**Heads-up for support:**
- **The Anthropic provider's "Claude Pro/Max" sign-in is removed.** A login saved with it no longer connects; Anthropic API keys are unaffected. The TUI shows a one-time notice, and a `--model anthropic/...` that no longer resolves says what to do. To keep using the plan: install Claude Code 2.1.296 or later, run `claude auth login`, then `/connect` → **Claude subscription**. The `claude-code-20250219` beta header is no longer sent.
- **Claude subscription needs `claude` on the same machine.** `OPENCODE_CLAUDE_PATH` points at a specific binary (on Windows, the full path to `claude.exe` if the login check does not finish). `claude` runs with the user's environment, so an `ANTHROPIC_API_KEY` or `ANTHROPIC_AUTH_TOKEN` in the shell is used, and billed, instead of the subscription. If no login is found within 5 minutes, `/connect` says to run `claude auth login`. Deleting a chat also deletes Claude Code's copy of it (`claude --resume` no longer lists it).
- **The first-run picker shows at most five suggestions:** the Altimate LLM Gateway, subscription logins found on this machine, the Claude and ChatGPT subscriptions, and the Anthropic and OpenAI API keys. Altimate Base, Google and every other provider are under **Search all providers…**. Detection only checks that a login exists; nothing is read from it or sent anywhere.
- **On a workspace with review on, a member's `skill publish` sends the skill for review** and exits 0; it reaches the team once the owner or an admin approves it in the Altimate web app. `altimate-code workspace review status` shows where each request stands.
- **Workspace admins review and manage like the owner.** Against an older server that sends no role, an admin is treated as a member.
- **Removing an approved skill or knowledge base from the workspace** (`/rules` → your request → **Remove from the workspace**) stops serving it to the whole team, and now asks for confirmation.
- **A capped agent (`steps`) ends with a text answer** sent with no tools, and the cap counts model requests rather than loop iterations, so compaction no longer uses up steps.

### Added

- **Claude subscription provider.** Claude models through the user's Claude Code login, with Altimate Code's tools, permissions, agents and skills; Claude Code's own tools, settings, `CLAUDE.md`, hooks and MCP servers are off. See "Claude Agent (Claude subscription)" in the providers docs. (#1706)
- **Subscription suggestions on the first-run picker and `/connect`**, from Claude Code, Codex CLI and Copilot editor logins and logins saved earlier, through `GET /altimate/provider/subscriptions`. (#1706)
- **Workspace review.** `skill publish` waits for approval on review-on workspaces, `workspace review status` lists your requests, and the TUI shows a notice once when a request is approved, rejected or removed. Skills sync at the version the workspace serves you. (#1685, #1708)
- **Workspace roles.** Admins approve, reject and manage like the owner, and their own submissions go in without review. (#1707)
- **Your requests in one view, from any client.** `/rules` lists your rule proposals and your skill and knowledge requests, and withdraws or removes them; `serve` gives the same through `GET /altimate/workspace/requests` and `POST /altimate/workspace/requests/{id}/withdraw|remove`, and `workspace review decisions --json` for plugins. (#1716)
- **Workspace rules in Claude Code and Codex sessions** through the plugins' hooks (`learn hook`, `learn rules`). (#1626)
- **Column answer in `impact_analysis`.** With a column and `direction` (`downstream` by default, or `upstream`), it traces the value hop by hop through compiled SQL and lists every model past the edited one as carrying it or possibly carrying it, with the reason (`metadata.column_answer`). `dbt_lineage` uses the same column resolver. See `impact_analysis` in the lineage tools docs. (#1710)
- **`tool report-event`**, which the Claude Code plugin calls to count how its lineage report is used (`lineage_report` telemetry; nothing is sent when telemetry is off). (#1718)

### Changed

- **The dbt integration moves to 0.3.17**, which starts its Python bridge without `python-bridge`, so the dbt-first path of `sql_execute` runs under Bun, and dbt's log lines stay off the TUI and JSON output. (#1368)

### Fixed

- **A capped agent ends with an answer** instead of a failed request, and a capped subagent no longer hands its parent an empty result; an empty task result is reported as such. (#1632)
- **One context overflow compacts once**, instead of replacing the summary with a summary of the kept tail, and the post-compaction ledger lists files written by shell commands. (#1639)
- **Background subagents run without blocking the parent** when `OPENCODE_EXPERIMENTAL_BACKGROUND_SUBAGENTS` is on: `background: true` returns a `task_id` to collect later, at most 4 run per parent, and cancellation keeps partial findings. (#1643)
- **Workspace telemetry.** A workspace's first prompt and job are sent from any session bound to it, with `origin`; a pinned IDE session names the picked workspace; a failed prompt leaves no workspace verdict; and `ALTIMATE_DISABLE_WORKSPACE` stops a saved link's first events. (#1699, #1672, #1705, #1711)
- **Clearer failures.** A failed `impact_analysis` column answer is said in the report, a Claude Code start that times out names the network or proxy as the likely cause, and a Claude subscription `/connect` that finds no login says to run `claude auth login`.

## [0.12.10] - 2026-11-07

**Workspace rules are no longer read from the repository, and a session's title no longer sticks on its placeholder.** The workspace rules state (queued proposals, pending approvals, review notices, dismissed topics and the personal learning choice) moves out of the project folder to a private per-user file, so a branch, merge or pull cannot plant a proposal or an approval that a client would then send with the signed-in person's credential, and team rules reach a session only for the account that pulled them. A session whose first title request fails now retries on the next messages and records the provider's error. Also in this release: a project linked by its path can gain its git remote so team rules can sync, `learn status` says why rules are off, and `tool` commands deliver a large answer whole on a pipe.
Expand Down
7 changes: 5 additions & 2 deletions docs/docs/configure/providers.md
Original file line number Diff line number Diff line change
Expand Up @@ -168,7 +168,7 @@ A login counts as found even if it has expired. Altimate Code checks that these
| Claude Pro/Max | Claude Code's login (`~/.claude/.credentials.json`, `~/.claude.json`, or the macOS keychain item `Claude Code-credentials`) | Uses that login; nothing to sign in |
| ChatGPT Plus/Pro | The Codex CLI's ChatGPT login (`~/.codex/auth.json`) | You sign in with ChatGPT once in Altimate Code |
| GitHub Copilot | A Copilot editor login (`~/.config/github-copilot/`) | You sign in with GitHub once in Altimate Code |
| Claude Pro/Max | Claude subscription connected earlier in Altimate Code | Checks your Claude Code login again; run `claude auth login` if it has ended |
| Claude subscription already connected | Claude subscription connected earlier in Altimate Code | Checks your Claude Code login again; run `claude auth login` if it has ended |
| ChatGPT, GitHub Copilot, SuperGrok or GitLab Duo | A login saved earlier in Altimate Code | Reconnects, signing in again if it expired |

## Claude Agent (Claude subscription)
Expand All @@ -187,7 +187,10 @@ How it behaves:
- Chats use Claude Code's system prompt plus the selected agent's prompt as its role. Temperature, top-p and top-k are not supported.
- You can switch between Claude Agent and other models in the same session. Claude is sent the part of the conversation it has not seen.
- Each session keeps one `claude` process, which stops after 5 minutes idle and resumes the saved Claude session on the next message.
- `claude` must be installed on the same machine. Set `OPENCODE_CLAUDE_PATH` to run a specific `claude` binary instead of the one on `PATH`.
- `claude` must be installed on the same machine. Set `OPENCODE_CLAUDE_PATH` to run a specific `claude` binary instead of the one on `PATH`, for example the full path to `claude.exe` on Windows if the login check does not finish.
- `claude` runs with your environment, as it does when you start it yourself. An `ANTHROPIC_API_KEY` or `ANTHROPIC_AUTH_TOKEN` set in your shell is therefore used instead of your subscription and billed to that key; unset it to use the subscription. Proxy settings (`HTTPS_PROXY`, and `NODE_EXTRA_CA_CERTS` for a company certificate) reach `claude` the same way.
- The model list comes from Claude Code, so it appears after the first successful connect. If `claude` cannot reach Anthropic, `/connect` keeps checking and then says no login was found.
- Deleting a chat in Altimate Code also deletes Claude Code's copy of that conversation, so it no longer appears in `claude --resume`. If that delete fails, the error is logged and the chat is still removed.

Connection settings in config map to Claude Code variables:

Expand Down
20 changes: 20 additions & 0 deletions docs/docs/data-engineering/tools/lineage-tools.md
Original file line number Diff line number Diff line change
Expand Up @@ -87,3 +87,23 @@ Source columns:

Downstream impact: dim_products.category rename would break fct_revenue
```

## impact_analysis

Shows what a change to a dbt model, or to one of its columns, affects downstream, from the project's `manifest.json`.

```
impact_analysis({ model: "stg_orders", column: "order_id", change_type: "rename" })
impact_analysis({ model: "fct_orders", column: "revenue", change_type: "modify", direction: "upstream" })
```

| Parameter | Meaning |
|---|---|
| `model` | The model, by name or unique id |
| `column` | Optional. Without it, the report is model-level only |
| `change_type` | The change you plan: `remove`, `rename`, `retype`, `add` or `modify` |
| `manifest_path` | Path to `manifest.json`; `target/manifest.json` by default |
| `direction` | With a column: `downstream` (default) says where its value goes; `upstream` traces where it comes from |
| `dialect` | Optional SQL dialect; read from the manifest when left out |

The report lists the directly and transitively affected models, their tests, and the exposures (dashboards) on them. With a column, it also traces each model's compiled SQL hop by hop and adds a column answer: the columns that carry the value, in which models, how many hops, and where it ends. A model the trace cannot settle (for example a `select *`, a natural join, or SQL the engine cannot read) counts as one that *may carry* it rather than being dropped. The report gives the counts; each model, with its reason, is in the tool's `metadata.column_answer`. If the column answer cannot be built, the report says why, and the model-level impact still stands.
4 changes: 4 additions & 0 deletions docs/docs/usage/tool.md
Original file line number Diff line number Diff line change
Expand Up @@ -80,3 +80,7 @@ The daemon listens on a local socket (a named pipe on Windows) under the Altimat
## Claude Code routing

For clients that cannot use the `altimate-code` MCP server, the Claude Code plugin adds CLI routing guidance at session start: `altimate-code tool context` followed by `altimate-code tool run` for data questions, and `altimate-code run` for long multi-step jobs. To persist the same guidance in the repository, run `altimate-code link --claude-md`; it writes a managed block to the project's `CLAUDE.md` and never changes content outside the `<!-- altimate-code:routing start -->` / `<!-- altimate-code:routing end -->` markers.

## Plugin events

`altimate-code tool report-event <action> --session claude:<id>` is called by the Claude Code plugin to record what a person did with its lineage report. `<action>` is one of `summary_shown`, `open_pressed`, `command_run`, `built`, `build_failed` or `no_browser`. It sends one `lineage_report` usage event (see [Telemetry](../reference/telemetry.md)) and nothing when telemetry is off. It is not meant to be run by hand.
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,10 @@ import type { AccountInfo, ModelInfo, Options, Query } from "@anthropic-ai/claud
import { Effect, Schema } from "effect"
import { ClaudeSDK } from "./sdk"

/** Why `claude` may not start in time; the probe and a chat turn both show it. */
export const START_TIMEOUT_MESSAGE =
"claude did not start within 30 seconds. Check the network, and any proxy (HTTPS_PROXY) or custom CA (NODE_EXTRA_CA_CERTS) it needs"

export interface Model extends ModelInfo {
contextWindow?: number
}
Expand Down Expand Up @@ -76,7 +80,7 @@ export const run = Effect.fn("ClaudeProbe.run")(function* (options: Options) {
}).pipe(
Effect.timeoutOrElse({
duration: "30 seconds",
orElse: () => Effect.fail(new ProbeError({ cause: new Error("claude did not start within 30 seconds") })),
orElse: () => Effect.fail(new ProbeError({ cause: new Error(START_TIMEOUT_MESSAGE) })),
}),
)
// No prompt is sent here, so an old Claude Code can still list its models.
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -20,6 +20,7 @@ import { ConvertToClaudeMessage } from "../language-model/convert-to-claude-mess
import { PendingTools } from "../tools/pending"
import { ToolNames } from "../tools/names"
import { ToolResult } from "../tools/result"
import { START_TIMEOUT_MESSAGE } from "./probe"

export const HANDOVER_MS = 10_000
export const IDLE_MS = 5 * 60_000
Expand Down Expand Up @@ -192,7 +193,7 @@ export function registry() {
attempt(() => process.session.ready).pipe(
Effect.timeoutOrElse({
duration: "30 seconds",
orElse: () => Effect.fail(new Error("claude did not start within 30 seconds")),
orElse: () => Effect.fail(new Error(START_TIMEOUT_MESSAGE)),
}),
),
).pipe(Effect.onError(() => stop(entry).pipe(Effect.ignore)))
Expand Down
10 changes: 10 additions & 0 deletions packages/opencode/src/altimate/provider/model-hint.ts
Original file line number Diff line number Diff line change
@@ -0,0 +1,10 @@
// altimate_change - new file
//
// Extra guidance for a "Model not found" error. The Anthropic provider lost its Claude Pro/Max sign-in
// in 0.12.11, so a saved `anthropic/...` model with only that login now fails to resolve; the bare
// error gave no reason. Only a missing provider gets the hint: an Anthropic provider that is there
// with an unknown model name needs a different model, not a key.
export function modelNotFoundHint(providerID: string | undefined, providerMissing: boolean | undefined) {
if (providerID !== "anthropic" || providerMissing !== true) return undefined
return "The Anthropic provider needs an API key. For a Claude Pro or Max plan, run /connect and pick Claude subscription."
}
7 changes: 6 additions & 1 deletion packages/opencode/src/altimate/tools/impact-analysis.ts
Original file line number Diff line number Diff line change
Expand Up @@ -136,7 +136,12 @@ export const ImpactAnalysisTool = Tool.define("impact_analysis", {
columnReach: reach,
columnReachError: reachError,
})
const output = answer ? `${report}\n${describeAnswer(answer)}` : report
// A failed column answer is said in the report, so it does not read as "no column impact". The same
// reason as column reach's is already in the report.
let output = report
if (answer) output = `${report}\n${describeAnswer(answer)}`
else if (answerError && answerError !== reachError)
output = `${report}\nColumn answer unavailable: ${answerError}`

const totalAffected = downstream.length
const severity =
Expand Down
2 changes: 1 addition & 1 deletion packages/opencode/src/altimate/workspace/skill-publish.ts
Original file line number Diff line number Diff line change
Expand Up @@ -1074,7 +1074,7 @@ export function describePublish(report: PublishReport): string {
return `Already waiting for review: "${name}" was sent to ${owner} with the same content. Nothing new was sent.`
if (review.duplicate) return `Already approved: "${name}" is in "${workspace}" with the same content. Nothing new was sent.`
if (review.status === "active") return `Published "${name}" to "${workspace}" ${detail}.`
return `Sent "${name}" for review to ${owner} ${detail}. It reaches the workspace once approved; \`altimate-code workspace review status\` shows where it stands.`
return `Sent "${name}" for review to ${owner} ${detail}. It reaches the workspace once approved in the Altimate web app; \`altimate-code workspace review status\` shows where it stands.`
}

/** Every error this module raises on purpose, with its stable code for clients that explain it in their own words
Expand Down
1 change: 1 addition & 0 deletions packages/opencode/src/cli/cmd/run.ts
Original file line number Diff line number Diff line change
Expand Up @@ -695,6 +695,7 @@ You are speaking to a non-technical business executive. Follow these rules stric
providerID: parsed.providerID,
modelID: parsed.modelID,
suggestions: provider ? Object.keys(provider.models).slice(0, 5) : [],
providerMissing: !provider,
})
}
}
Expand Down
9 changes: 8 additions & 1 deletion packages/opencode/src/cli/error.ts
Original file line number Diff line number Diff line change
@@ -1,6 +1,8 @@
import { NamedError } from "@opencode-ai/core/util/error"
import { errorFormat } from "@/util/error"
import { isRecord } from "@/util/record"
// altimate_change — explain a missing Anthropic provider (Claude Pro/Max sign-in removed)
import { modelNotFoundHint } from "@/altimate/provider/model-hint"

type ConfigIssue = { message: string; path: string[] }

Expand Down Expand Up @@ -57,16 +59,21 @@ export function FormatError(input: unknown): string | undefined {
return stringField(input, "message") ?? ""
}

// ProviderModelNotFoundError: { providerID: string, modelID: string, suggestions?: string[] }
// ProviderModelNotFoundError: { providerID: string, modelID: string, suggestions?: string[], providerMissing?: boolean }
const providerModelNotFound = configData(input, "ProviderModelNotFoundError")
if (providerModelNotFound) {
const suggestions = Array.isArray(providerModelNotFound.suggestions)
? providerModelNotFound.suggestions.filter((x) => typeof x === "string")
: []
// altimate_change start — upstream_fix: branding regressions in user-visible error strings
const hint = modelNotFoundHint(
stringField(providerModelNotFound, "providerID"),
providerModelNotFound.providerMissing === true,
)
return [
`Model not found: ${stringField(providerModelNotFound, "providerID")}/${stringField(providerModelNotFound, "modelID")}`,
...(suggestions.length ? ["Did you mean: " + suggestions.join(", ")] : []),
...(hint ? [hint] : []),
`Try: \`altimate models\` to list available models`,
`Or check your config (altimate-code.json) provider/model names`,
].join("\n")
Expand Down
Loading
Loading