c49906ec77
fix(coding-agent): preserve managed state file permissions
...
Closes #7779
2026-08-22 01:19:25 +02:00
f4585b8bec
fix(coding-agent): simplify session sharing links
2026-08-21 18:33:27 +02:00
77f2d1235e
chore(interactive-mode): get rid of theme, only share via radius if logged in
2026-08-21 16:47:08 +02:00
a2f369d63a
fix(slash-commands): order tree above thinking
2026-08-21 16:14:10 +02:00
686f3487f5
feat(interactive-mode): share via radius artifacts under experimental ( #8443 )
2026-08-21 16:05:01 +02:00
5cd93f688a
feat(coding-agent): add development pi wrapper
2026-08-20 15:59:38 +02:00
8f2ae3fadd
fix(tui): prevent wrapped table link color leaks ( #8363 )
2026-08-20 15:58:47 +02:00
cffe4d776c
fix(settings-selector): nit ordering in default t.l. per model
2026-08-20 14:42:47 +02:00
768184923a
fix(settings-selector): nit spacing
2026-08-20 14:28:33 +02:00
1d3503fb9b
feat(settings-selector): show default, make default searchable for model and thinking ( #8399 )
2026-08-20 13:55:27 +02:00
5b3caaf4cb
chore(settings-selector): get rid of default thinking in settings, ctrl + S is enough
2026-08-20 13:35:21 +02:00
5133c9284f
chore(settings-selector): get rid of --default and global model
...
ctrl+s is enough, there goes my joy
2026-08-20 13:30:47 +02:00
b7bb00b936
fix(ai): retain reasoning details in thinking signature
2026-08-19 23:35:43 +02:00
496185f6e4
feat(coding-agent): /thinking command
2026-08-19 18:54:52 +02:00
9c8070fbe4
feat(settings-selector): ctrl + s persists /model
2026-08-19 18:41:53 +02:00
f0a2880f29
fix(settings-selector): revert token estimate removal
2026-08-19 18:26:36 +02:00
a669db3c33
fix(settings-selector): show modelid [provider] like /model
2026-08-19 18:25:14 +02:00
ee29aa118b
feat(settings-selector): make default model and thinking level searchable
2026-08-19 17:43:21 +02:00
98767a25d2
fix(settings-selector): remove token estimates
2026-08-19 17:29:02 +02:00
2ff8ba6223
fix(coding-agent): keep model and thinking level changes session scoped ( #8356 )
...
* fix(coding-agent): keep model and thinking level changes session scoped
* fix(coding-agent): persist requested default thinking level
* feat: --default arg to persist
* fix
* fix(coding-agent): keep model and thinking level changes session scoped
* fix: make default thinking level per model show list of models
* cleanup export
2026-08-19 16:48:29 +02:00
4ca636c5e0
feat(ai): opeani completions reasoning details ( #8246 )
...
* fix(ai): opeani completions reasoning details
* cleanup: type i didnt like
* changelog
2026-08-19 16:44:17 +02:00
87af49dec2
Add pi user-agent to most api adapters ( #8361 )
...
added Pi’s default User-Agent to seven adapters:
- openai-responses
- openai-completions
- anthropic-messages
- azure-openai-responses
- google-generative-ai
- google-vertex
- mistral-conversations
Closes #8305
2026-08-19 15:56:26 +02:00
1355cd36e0
fix(coding-agent): normalize UTF-8 BOMs in text inputs
...
closes #8337
2026-08-19 14:11:02 +02:00
4495469a5e
fix(coding-agent): compact without provider usage
...
closes #8328
2026-08-19 13:44:22 +02:00
3de00332f7
fix(ai): derive Z.AI reasoning effort metadata
...
Use models.dev effort options for Z.AI models while preserving GLM-5.2's none effort. Expose GLM-5.3's low, high, and max levels.
closes #8336
2026-08-19 13:19:52 +02:00
678f0af30d
fix(coding-agent): show startup diagnostics in TUI
...
fixes #7829
2026-08-19 13:07:57 +02:00
913bcf3391
fix(coding-agent): report settings diagnostic paths
2026-08-19 13:07:53 +02:00
1e1a6e27be
feat(coding-agent): include paths in settings errors
2026-08-19 13:07:20 +02:00
d57e531f5d
fix(ai): round-trip Bedrock redacted reasoning ( #8314 )
...
Bedrock Converse returns encrypted reasoning from non-Anthropic models
(e.g. OpenAI GPT-5.6) as the opaque `reasoningContent.redactedContent`
member. The stream handler only read `text` and `signature`, so the
payload was dropped and replayed history lost the reasoning block.
Buffer the delta bytes, encode them into `thinkingSignature` with
`redacted: true` (the representation the Anthropic path already uses),
and lower them back to `reasoningContent.redactedContent` on the next
turn. Scratch state is stripped from every terminal path, since a
stream can settle without stopping each block.
2026-08-19 12:42:27 +02:00
830a0a59e9
fix(coding-agent): expose tool metadata at stream start ( #7953 )
...
* fix(coding-agent): expose tool metadata at stream start
* fix(coding-agent): centralize message update serialization
2026-08-19 12:24:30 +02:00
d711bd5f0a
fix(coding-agent): preserve branch summary source leaf
...
Record the pre-navigation leaf in branch summary fromId instead of the destination node.
2026-08-19 12:02:37 +02:00
ed867e9094
fix(ai): fallback cost not via stream options ( #8352 )
...
* fix(ai): fallback cost not via stream options
* fix: always pass beta header, remove fallback from stream options
2026-08-19 11:30:26 +02:00
3a0b9a3eee
Revert "feat(agent): expose provider context construction"
...
This reverts commit 2509b5c037
2026-08-19 09:53:15 +02:00
8dab70281b
Revert "feat(coding-agent): add cache-friendly compaction primitives"
...
This reverts commit cff1cf52c6
2026-08-19 09:52:26 +02:00
55b0db4d3e
fix(ai): prevent copilot policy login rate limits ( #8254 )
...
- fixes #7850
- fetch the account model catalog before policy updates
- update only known, tool-capable, unconfigured models
- retry throttled login requests within a bounded delay
- preserve the existing non-retrying refresh behavior
2026-08-19 09:40:18 +02:00
4809c2abca
fix(ai): anthropic fallback usage ( #8319 )
...
* fix: anthropic fallback usage
* small cleanup
2026-08-19 09:11:45 +02:00
59a71b235d
Revert "fix(ai): anthropic fallback usage ( #8308 )" ( #8313 )
...
This reverts commit a6c6f80180
2026-08-18 17:10:35 +02:00
f0c5d86d20
fix(tui): fit text padding to narrow widths, closes #8252
2026-08-18 16:49:14 +02:00
c061328981
fix(coding-agent): load extensions in Node SEA hosts, closes #8237
2026-08-18 16:48:18 +02:00
b237412699
feat(ai): generalize openai-completions thinking token budget fields ( #8275 )
...
#7638 hardcoded vLLM's thinking_token_budget. Map thinkingBudgets through
thinkingTokenBudgetField, keep the boolean as a vLLM alias, clamp so the
answer keeps 1024 tokens, and document it.
fixes #8274
Co-authored-by: Cursor <cursoragent@cursor.com >
Co-authored-by: Mario Zechner <badlogicgames@gmail.com >
2026-08-18 16:35:35 +02:00
a6c6f80180
fix(ai): anthropic fallback usage ( #8308 )
...
* fix(ai): anthropic fallback usage
* changeloggg
2026-08-18 15:28:12 +02:00
cff1cf52c6
feat(coding-agent): add cache-friendly compaction primitives
...
Add provider-context prefix preparation and optional summary request settings needed by cache-friendly compaction. Existing AgentSession callers continue to pass undefined, so standalone summarization remains the default until the feature is wired in.
2026-08-18 14:52:06 +02:00
ef8dc7385b
feat(coding-agent): centralize compaction summary requests
2026-08-18 14:19:01 +02:00
836aee6d38
feat(coding-agent): show compaction usage notices
2026-08-18 13:27:00 +02:00
2509b5c037
feat(agent): expose provider context construction
2026-08-18 09:34:47 +02:00
8af7690c4f
fix(coding-agent): skip trusted subagent prompts
...
closes #8261
2026-08-18 09:26:07 +02:00
90305d90a0
fix(coding-agent): disable tools during summarization
2026-08-18 09:19:40 +02:00
e5dde9a76b
feat(ai): add simple tool choice option
2026-08-18 08:41:24 +02:00
ad58801ce7
fix(ai): update Baseten GLM input modalities
2026-08-18 08:19:46 +02:00
9117326b4c
fix(ai): forward Azure Responses tool choice
2026-08-18 07:56:51 +02:00
Armin Ronacher