Skip to content

armgit / eve

Public Imported from GitHub
0

The Open Framework for Building Agents

Version Packages (#3396)

github-actions[bot]4c3b93356aa1
diff --git a/.changeset/agent-tool-reexport-dispatch.md b/.changeset/agent-tool-reexport-dispatch.md
deleted file mode 100644
index 856e9d0c9..000000000
--- a/.changeset/agent-tool-reexport-dispatch.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"eve": patch
----
-
-Restored `agent` tools now dispatch through the shared subagent workflow when framework defaults are disabled.
diff --git a/.changeset/compact-model-footer.md b/.changeset/compact-model-footer.md
deleted file mode 100644
index 941fd9bea..000000000
--- a/.changeset/compact-model-footer.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"eve": patch
----
-
-Simplify the TUI footer by removing the local port and separating the model and connection with dots. Vercel account connections show the resolved team slug instead of a team ID.
diff --git a/.changeset/fair-services-build.md b/.changeset/fair-services-build.md
deleted file mode 100644
index 689956a23..000000000
--- a/.changeset/fair-services-build.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"eve": patch
----
-
-Declare `.vercel/output` for generated eve services so Vercel consumes the Build Output API directory that their build commands produce instead of looking for the framework preset's `.output` default.
diff --git a/.changeset/faster-model-connection.md b/.changeset/faster-model-connection.md
deleted file mode 100644
index c3ef2ef9b..000000000
--- a/.changeset/faster-model-connection.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"eve": patch
----
-
-Model login shares resolved credentials with the running agent, avoids rebuilds for account, team, and key changes, and refreshes agent information without delaying chat. Clear status messages show when to finish browser sign-in and what eve is waiting for, without flashing indicators for quick steps.
diff --git a/.changeset/final-activity-flush.md b/.changeset/final-activity-flush.md
deleted file mode 100644
index 42c8efdfb..000000000
--- a/.changeset/final-activity-flush.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"eve": patch
----
-
-Flush the final activity snapshot when an activity stream closes during the render debounce window.
diff --git a/.changeset/fresh-vercels-sleep.md b/.changeset/fresh-vercels-sleep.md
deleted file mode 100644
index e8ff87aa3..000000000
--- a/.changeset/fresh-vercels-sleep.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"eve": patch
----
-
-Make `withEve` discover a containing workspace when Vercel evaluates `vercel.ts` from its temporary `.vercel` directory.
diff --git a/.changeset/gateway-object-prompt-caching.md b/.changeset/gateway-object-prompt-caching.md
deleted file mode 100644
index cc62b11c9..000000000
--- a/.changeset/gateway-object-prompt-caching.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"eve": patch
----
-
-Keep automatic prompt caching enabled when local model authentication resolves an AI Gateway model to a provider object. Anthropic models retain cache reuse after signing in through `eve dev`.
diff --git a/.changeset/leased-session-streams.md b/.changeset/leased-session-streams.md
deleted file mode 100644
index afaa5cb69..000000000
--- a/.changeset/leased-session-streams.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"eve": patch
----
-
-Bound negotiated session stream responses with renewable leases so abandoned serverless invocations release their durable stream readers. The eve client renews these responses from its cursor without exposing transport heartbeats or lease records as session events; tail-relative reads and streams with reconnection disabled remain unleased.
diff --git a/.changeset/local-workflow-health-probe.md b/.changeset/local-workflow-health-probe.md
deleted file mode 100644
index ca9f66ca0..000000000
--- a/.changeset/local-workflow-health-probe.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"eve": patch
----
-
-Fix a local workflow health-check failure after model or source changes. Capability probes no longer look up a run before it exists, avoiding a spurious queue error and startup delay.
diff --git a/.changeset/login-team-switch.md b/.changeset/login-team-switch.md
deleted file mode 100644
index 38aca413c..000000000
--- a/.changeset/login-team-switch.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"eve": patch
----
-
-Allow switching Vercel teams through `/login` by always showing the team picker when multiple teams are available. The current team is highlighted rather than silently selected; automatic startup still reuses the saved connection.
diff --git a/.changeset/muted-vercel-footer.md b/.changeset/muted-vercel-footer.md
deleted file mode 100644
index 5fd2379d6..000000000
--- a/.changeset/muted-vercel-footer.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"eve": patch
----
-
-Match the Vercel label's color to the surrounding model and team text in the TUI footer.
diff --git a/.changeset/offer-confirmed-self-mod-repairs.md b/.changeset/offer-confirmed-self-mod-repairs.md
deleted file mode 100644
index 298b476b9..000000000
--- a/.changeset/offer-confirmed-self-mod-repairs.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"eve": patch
----
-
-When a self-modified tool or capability later fails or behaves incorrectly, the agent now offers to delegate a repair. It waits for your confirmation before sending the observed and expected behavior back to self-modification.
diff --git a/.changeset/prominent-empty-prompt.md b/.changeset/prominent-empty-prompt.md
deleted file mode 100644
index 7d81d9f62..000000000
--- a/.changeset/prominent-empty-prompt.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"eve": patch
----
-
-Use the same full-size arrow for empty and populated TUI prompts, with the empty arrow in the terminal's default foreground color for better visibility.
diff --git a/.changeset/quiet-model-login.md b/.changeset/quiet-model-login.md
deleted file mode 100644
index ddefb4798..000000000
--- a/.changeset/quiet-model-login.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"eve": patch
----
-
-Open the TUI directly after init, connect models through `/login` without requiring a Vercel project, and store local credentials securely. Add direct OpenAI and Anthropic model helpers, apply model selections immediately, and simplify menus and `/add` to get to chat sooner. Vercel CLI installation and login now run during deployment instead of separate `/vc:install` and `/vc:login` commands.
diff --git a/.changeset/recognize-gateway-budget-refusals.md b/.changeset/recognize-gateway-budget-refusals.md
deleted file mode 100644
index 5776ca2ef..000000000
--- a/.changeset/recognize-gateway-budget-refusals.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"eve": patch
----
-
-Recognize AI Gateway credit-balance and project-budget refusals in semantic error summaries.
diff --git a/.changeset/remote-trusted-sources-startup.md b/.changeset/remote-trusted-sources-startup.md
deleted file mode 100644
index d37886661..000000000
--- a/.changeset/remote-trusted-sources-startup.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"eve": patch
----
-
-Restore guided Trusted Sources setup when Deployment Protection blocks `eve dev --url`. eve reuses the existing Vercel CLI session, asks before changing project access, and verifies the connection before returning to chat.
diff --git a/.changeset/sandbox-release-image.md b/.changeset/sandbox-release-image.md
deleted file mode 100644
index 4f8ea7425..000000000
--- a/.changeset/sandbox-release-image.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"eve": patch
----
-
-Use the base release version for default sandbox images when running commit or git-ref tarballs. Build metadata such as `+git.<sha>` no longer produces invalid image tags for Docker, microsandbox, or Vercel Sandbox.
diff --git a/.changeset/search-add-capabilities.md b/.changeset/search-add-capabilities.md
deleted file mode 100644
index 953d90b72..000000000
--- a/.changeset/search-add-capabilities.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"eve": patch
----
-
-Search `/add` by capabilities such as iMessage or SMS, as well as item names and addresses. The picker now searches catalog titles and descriptions without adding extra text to menu rows.
diff --git a/.changeset/self-modification-vercel-scope.md b/.changeset/self-modification-vercel-scope.md
deleted file mode 100644
index 15755455c..000000000
--- a/.changeset/self-modification-vercel-scope.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"eve": patch
----
-
-Prompt for and resolve the target Vercel project before configuring production self-modification, and scope GitHub connector creation and attachment to that project's owner.
diff --git a/.changeset/settled-startup-presentation.md b/.changeset/settled-startup-presentation.md
deleted file mode 100644
index 4929e3aee..000000000
--- a/.changeset/settled-startup-presentation.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"eve": patch
----
-
-Keep the composer available while startup connects your model and prepares chat, with animated progress and prompts only when input is needed. Preserve drafts and queued messages through setup, and announce readiness only after the runtime refresh succeeds.
diff --git a/.changeset/short-stream-bursts.md b/.changeset/short-stream-bursts.md
deleted file mode 100644
index d90f79adb..000000000
--- a/.changeset/short-stream-bursts.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"eve": patch
----
-
-Batch closely spaced stream events on Vercel within a 10 ms window to reduce network writes. Explicit Workflow stream-flush overrides remain supported.
diff --git a/.changeset/single-owner-sessions.md b/.changeset/single-owner-sessions.md
deleted file mode 100644
index 04ef750a9..000000000
--- a/.changeset/single-owner-sessions.md
+++ /dev/null
@@ -1,9 +0,0 @@
----
-"eve": minor
----
-
-Run every conversational turn directly inside the session's owning workflow instead of dispatching a child turn run per message. An idle session hands its settled state to the exact deployment that accepted a new delivery, whether the delivery arrives through the session ID or any channel continuation address, keeping the original session ID and event stream and renewing the session's configured timeout. Deliveries that land while a handoff is in progress wait for the successor instead of starting a replacement session.
-
-Steering a running turn applies at the next committed step boundary without cancelling in-flight model or tool work and preserves the turn's identity and usage; input that arrives after the model has answered starts the next turn. `continuation.rekey()` is replaced by `continuation.alias()`: every claimed address stays active, and the most recently selected alias is exposed as `continuation.token`.
-
-Sessions from the former driver/turn execution model are imported on their next turn, preserving identity, history, and the original stream while interrupting pending work; drivers started before eve 0.45 are reported inactive and their channel starts a fresh session. Retain the original deployment until imported sessions end, and retire sessions before rolling back across this boundary.
diff --git a/.changeset/spotty-traces-follow.md b/.changeset/spotty-traces-follow.md
deleted file mode 100644
index b4dc1a975..000000000
--- a/.changeset/spotty-traces-follow.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"eve": patch
----
-
-Emits canonical GenAI cache-write tokens, reports the provider-confirmed response model, and adds the executing agent name to tool spans.
diff --git a/.changeset/stable-prompt-hint.md b/.changeset/stable-prompt-hint.md
deleted file mode 100644
index d723c16e8..000000000
--- a/.changeset/stable-prompt-hint.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"eve": patch
----
-
-Replace rotating TUI prompt suggestions with a stable “Send a message…” placeholder that does not assume agent capabilities.
diff --git a/.changeset/steer-before-turn-completion.md b/.changeset/steer-before-turn-completion.md
deleted file mode 100644
index dd37b5bd7..000000000
--- a/.changeset/steer-before-turn-completion.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"eve": patch
----
-
-Apply steering in the core runtime before a pending model request produces an answer, continuing the same turn with the correction across clients and channels. Preserve executing tools and their results, prevent duplicate in-process model execution on background wakeups, and keep boundary steering after assistant output begins.
diff --git a/.changeset/steer-on-enter.md b/.changeset/steer-on-enter.md
deleted file mode 100644
index 4d6070a1f..000000000
--- a/.changeset/steer-on-enter.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"eve": patch
----
-
-Pressing Enter during a TUI response now steers the active agent immediately instead of queuing the message until the turn finishes. Startup messages and sessions without steering support continue to queue.
diff --git a/.changeset/steering-gutter-color.md b/.changeset/steering-gutter-color.md
deleted file mode 100644
index 13cd6c23a..000000000
--- a/.changeset/steering-gutter-color.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"eve": patch
----
-
-Distinguish steered TUI messages with a yellow left gutter instead of an extra arrow row. Queued messages also render without an arrow row.
diff --git a/.changeset/tidy-skills-announce.md b/.changeset/tidy-skills-announce.md
deleted file mode 100644
index fd78f9d35..000000000
--- a/.changeset/tidy-skills-announce.md
+++ /dev/null
@@ -1,5 +0,0 @@
----
-"eve": patch
----
-
-Announce when dynamic skills are withdrawn completely so agents do not retain stale skill lists.
diff --git a/packages/eve-buzz-acp-adapter/CHANGELOG.md b/packages/eve-buzz-acp-adapter/CHANGELOG.md
index b65dfca96..fcec49a35 100644
--- a/packages/eve-buzz-acp-adapter/CHANGELOG.md
+++ b/packages/eve-buzz-acp-adapter/CHANGELOG.md
@@ -1,5 +1,39 @@
# @eve/buzz-acp-adapter
+## 0.0.28
+
+### Patch Changes
+
+- Updated dependencies [32aca9b]
+- Updated dependencies [7b17b27]
+- Updated dependencies [f21e63f]
+- Updated dependencies [7b17b27]
+- Updated dependencies [e1ae8eb]
+- Updated dependencies [f21e63f]
+- Updated dependencies [fe6451c]
+- Updated dependencies [032860c]
+- Updated dependencies [7b17b27]
+- Updated dependencies [7b17b27]
+- Updated dependencies [7b17b27]
+- Updated dependencies [ba0b866]
+- Updated dependencies [7b17b27]
+- Updated dependencies [7b17b27]
+- Updated dependencies [8bb62e2]
+- Updated dependencies [7b17b27]
+- Updated dependencies [7b17b27]
+- Updated dependencies [7b17b27]
+- Updated dependencies [a08e4bb]
+- Updated dependencies [7b17b27]
+- Updated dependencies [61d4097]
+- Updated dependencies [61d4097]
+- Updated dependencies [53fad6c]
+- Updated dependencies [7b17b27]
+- Updated dependencies [fe6451c]
+- Updated dependencies [7b17b27]
+- Updated dependencies [7b17b27]
+- Updated dependencies [1e6437b]
+
## 0.0.27
### Patch Changes
diff --git a/packages/eve-buzz-acp-adapter/package.json b/packages/eve-buzz-acp-adapter/package.json
index c47ce3693..407d82be9 100644
--- a/packages/eve-buzz-acp-adapter/package.json
+++ b/packages/eve-buzz-acp-adapter/package.json
@@ -1,6 +1,6 @@
{
"name": "@eve/buzz-acp-adapter",
- "version": "0.0.27",
+ "version": "0.0.28",
"description": "Buzz ACP compatibility adapter for eve agents.",
"keywords": [
"acp",
diff --git a/packages/eve/CHANGELOG.md b/packages/eve/CHANGELOG.md
index 5c0ea23cd..66db614a5 100644
--- a/packages/eve/CHANGELOG.md
+++ b/packages/eve/CHANGELOG.md
@@ -1,5 +1,45 @@
# eve
+## 0.57.0
+
+### Minor Changes
+
+- 61d4097: Run every conversational turn directly inside the session's owning workflow instead of dispatching a child turn run per message. An idle session hands its settled state to the exact deployment that accepted a new delivery, whether the delivery arrives through the session ID or any channel continuation address, keeping the original session ID and event stream and renewing the session's configured timeout. Deliveries that land while a handoff is in progress wait for the successor instead of starting a replacement session.
+
+ Steering a running turn applies at the next committed step boundary without cancelling in-flight model or tool work and preserves the turn's identity and usage; input that arrives after the model has answered starts the next turn. `continuation.rekey()` is replaced by `continuation.alias()`: every claimed address stays active, and the most recently selected alias is exposed as `continuation.token`.
+
+ Sessions from the former driver/turn execution model are imported on their next turn, preserving identity, history, and the original stream while interrupting pending work; drivers started before eve 0.45 are reported inactive and their channel starts a fresh session. Retain the original deployment until imported sessions end, and retire sessions before rolling back across this boundary.
+
+### Patch Changes
+
+- 32aca9b: Restored `agent` tools now dispatch through the shared subagent workflow when framework defaults are disabled.
+- 7b17b27: Simplify the TUI footer by removing the local port and separating the model and connection with dots. Vercel account connections show the resolved team slug instead of a team ID.
+- f21e63f: Declare `.vercel/output` for generated eve services so Vercel consumes the Build Output API directory that their build commands produce instead of looking for the framework preset's `.output` default.
+- 7b17b27: Model login shares resolved credentials with the running agent, avoids rebuilds for account, team, and key changes, and refreshes agent information without delaying chat. Clear status messages show when to finish browser sign-in and what eve is waiting for, without flashing indicators for quick steps.
+- e1ae8eb: Flush the final activity snapshot when an activity stream closes during the render debounce window.
+- f21e63f: Make `withEve` discover a containing workspace when Vercel evaluates `vercel.ts` from its temporary `.vercel` directory.
+- fe6451c: Keep automatic prompt caching enabled when local model authentication resolves an AI Gateway model to a provider object. Anthropic models retain cache reuse after signing in through `eve dev`.
+- 032860c: Bound negotiated session stream responses with renewable leases so abandoned serverless invocations release their durable stream readers. The eve client renews these responses from its cursor without exposing transport heartbeats or lease records as session events; tail-relative reads and streams with reconnection disabled remain unleased.
+- 7b17b27: Fix a local workflow health-check failure after model or source changes. Capability probes no longer look up a run before it exists, avoiding a spurious queue error and startup delay.
+- 7b17b27: Allow switching Vercel teams through `/login` by always showing the team picker when multiple teams are available. The current team is highlighted rather than silently selected; automatic startup still reuses the saved connection.
+- 7b17b27: Match the Vercel label's color to the surrounding model and team text in the TUI footer.
+- ba0b866: When a self-modified tool or capability later fails or behaves incorrectly, the agent now offers to delegate a repair. It waits for your confirmation before sending the observed and expected behavior back to self-modification.
+- 7b17b27: Use the same full-size arrow for empty and populated TUI prompts, with the empty arrow in the terminal's default foreground color for better visibility.
+- 7b17b27: Open the TUI directly after init, connect models through `/login` without requiring a Vercel project, and store local credentials securely. Add direct OpenAI and Anthropic model helpers, apply model selections immediately, and simplify menus and `/add` to get to chat sooner. Vercel CLI installation and login now run during deployment instead of separate `/vc:install` and `/vc:login` commands.
+- 8bb62e2: Recognize AI Gateway credit-balance and project-budget refusals in semantic error summaries.
+- 7b17b27: Restore guided Trusted Sources setup when Deployment Protection blocks `eve dev --url`. eve reuses the existing Vercel CLI session, asks before changing project access, and verifies the connection before returning to chat.
+- 7b17b27: Use the base release version for default sandbox images when running commit or git-ref tarballs. Build metadata such as `+git.<sha>` no longer produces invalid image tags for Docker, microsandbox, or Vercel Sandbox.
+- 7b17b27: Search `/add` by capabilities such as iMessage or SMS, as well as item names and addresses. The picker now searches catalog titles and descriptions without adding extra text to menu rows.
+- a08e4bb: Prompt for and resolve the target Vercel project before configuring production self-modification, and scope GitHub connector creation and attachment to that project's owner.
+- 7b17b27: Keep the composer available while startup connects your model and prepares chat, with animated progress and prompts only when input is needed. Preserve drafts and queued messages through setup, and announce readiness only after the runtime refresh succeeds.
+- 61d4097: Batch closely spaced stream events on Vercel within a 10 ms window to reduce network writes. Explicit Workflow stream-flush overrides remain supported.
+- 53fad6c: Emits canonical GenAI cache-write tokens, reports the provider-confirmed response model, and adds the executing agent name to tool spans.
+- 7b17b27: Replace rotating TUI prompt suggestions with a stable “Send a message…” placeholder that does not assume agent capabilities.
+- fe6451c: Apply steering in the core runtime before a pending model request produces an answer, continuing the same turn with the correction across clients and channels. Preserve executing tools and their results, prevent duplicate in-process model execution on background wakeups, and keep boundary steering after assistant output begins.
+- 7b17b27: Pressing Enter during a TUI response now steers the active agent immediately instead of queuing the message until the turn finishes. Startup messages and sessions without steering support continue to queue.
+- 7b17b27: Distinguish steered TUI messages with a yellow left gutter instead of an extra arrow row. Queued messages also render without an arrow row.
+- 1e6437b: Announce when dynamic skills are withdrawn completely so agents do not retain stale skill lists.
+
## 0.56.0
### Minor Changes
diff --git a/packages/eve/package.json b/packages/eve/package.json
index 624233f9e..5a23905b8 100644
--- a/packages/eve/package.json
+++ b/packages/eve/package.json
@@ -1,6 +1,6 @@
{
"name": "eve",
- "version": "0.56.0",
+ "version": "0.57.0",
"private": false,
"description": "Filesystem-first framework for durable backend AI agents that run anywhere.",
"keywords": [