Introduces DefaultAgentHandle class that auto-creates a chat conversation
(POST /api/v1/assistants/chatConversations) when no conversation_id is
supplied, then calls /v1/agents/invoke with agentType=PLANNER and
agentMode=CHAT_MODE, mirroring the chat UI flow.

Also adds test_default_agent.py with two tests that currently fail because
client.agent() still requires a name argument — wiring is deferred to Task 6.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Mirrors the chat UI flow: auto-creates a chat conversation via
POST /api/v1/assistants/chatConversations, then calls the new
/v1/agents/invoke (sync) or /v1/agents/run (SSE) endpoints with
agentType=PLANNER, agentMode=CHAT_MODE.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Mirrors the Python SDK and the chat UI flow: auto-creates a chat
conversation via POST /api/v1/assistants/chatConversations, then calls
/v1/agents/invoke (sync) or /v1/agents/run (SSE) with agentType=PLANNER,
agentMode=CHAT_MODE.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Mirrors the Python and TypeScript SDKs and the chat UI flow:
auto-creates a chat conversation via POST /api/v1/assistants/chatConversations,
then calls /api/v1/agents/invoke (sync) or /api/v1/agents/run (SSE) with
agentType=PLANNER, agentMode=CHAT_MODE.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Mirrors the SDK pattern: auto-create chat conversation, then call
/v1/agents/invoke (sync) or /v1/agents/run (SSE) with PLANNER/CHAT_MODE.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

When enabled, the node uses the platform's default agent (PLANNER /
CHAT_MODE) instead of requiring an agent name. Mirrors client.agent()
(no-arg) on the TypeScript SDK.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Add examples of the no-arg client.agent() path (PLANNER / CHAT_MODE) to
the top-level CLAUDE.md API Pattern block and all per-SDK READMEs that
contain a quick-start example.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

The Use Default Agent toggle requires client.agent() (no-arg) which
landed in the TS SDK at 0.1.2. Pinning ^0.1.1 would let a fresh
install resolve the older SDK at runtime and break the new path.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Backend now separates Sender.SYSTEM (thinking) and Sender.ASSISTANT
(final body) messages in the sync invoke flow, exposing them as a new
thinkingSteps field on InvokeAgentResponse. Surfaced as
thinking_steps (Python, Rust) / thinkingSteps (TS, Java, n8n) on the
SDK's InvokeResponse model. Backward-compatible: defaults to [] when
the backend doesn't supply the field.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Empty stubs for AgentsAPI, BotsAPI, PersonasAPI, AbilitiesAPI, and
MemoriesAPI. Methods land in subsequent commits. __init__.py is a
docstring-only module — callers import from canonical paths.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

…odels

Adds memory models to ai_sdk.models:
- MemoryType, MemoryScope, MemoryVisibility enums
- CreateContextMemoryRequest with to_api_dict() that flattens
  visibility -> shareConfig.visibility on the wire
- ContextMemory.from_dict() that flattens shareConfig.visibility
  back to a top-level visibility field for ergonomics
- MemorySearchHit / MemorySearchResults parsed from OpenSearch shape

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Adds DELETE method to HTTPClient (sync + async) for soft/hard delete.
Implements MemoriesAPI with sync and async variants of every operation.
Search uses the NLQ hybrid endpoint at /api/v1/hybrid/nlq/search with
filters serialized as JSON query params.

The MemoryList type alias works around the type checker conflating the
list method with the builtin list type.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

…classes

AgentsAPI, BotsAPI, PersonasAPI, AbilitiesAPI each holds its own HTTP
clients and exposes list/get/create/etc. AgentsAPI takes a back-reference
to AISdk so create() can resolve persona and ability names via the new
namespaces.

Pagination helper is duplicated per file to keep each namespace
self-contained — extract if a fifth namespace appears.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Rewrites AISdk to expose:
- client.agents (list, create) + async variants
- client.bots (list, get) + async variants
- client.personas (list, get, create) + async variants
- client.abilities (list, get) + async variants
- client.memories (list, get, create, delete, search) + async variants

Removes: client.list_agents/create_agent, client.list_bots/get_bot,
client.list_personas/get_persona/create_persona, client.list_abilities/
get_ability, plus all aXxx async variants.

client.agent(name) is unchanged — it returns an AgentHandle.
client.mcp is unchanged.

Updates langchain integration code and all unit tests to use the new
namespace API. Modernizes 3 persona async-error tests from the
deprecated asyncio.get_event_loop() pattern to asyncio.run().

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

- Adds memory models to package-level __all__ and __init__.py
  re-exports so users can `from ai_sdk import ContextMemory, ...`
- Adds integration test that round-trips create -> get -> list ->
  search -> delete against a real instance (gated on AI_SDK_HOST
  and AI_SDK_TOKEN)

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

- Adds CHANGELOG.md with the breaking-change migration table for all
  four SDKs (Python, TypeScript, Java, Rust CLI) and the new
  client.memories namespace.
- Updates CLAUDE.md API Pattern section with namespace examples.
- Updates docs/quickstart.md, docs/async.md, docs/standalone.md to use
  client.<entity>.<verb>() form.
- Updates cookbook agent-config docs (gdpr-dsar-compliance, dbt-pr-review,
  dq-failure-slack-notifications).
- Updates n8n integration test to use client.agents.list().

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Adds list/get/create/delete/search subcommands under `ai-sdk memories`,
mirroring the Phase 1 Python MemoriesAPI surface. Hits
/api/v1/contextCenter/memories for CRUD and /api/v1/hybrid/nlq/search for
hybrid NLQ search over the contextMemory index.

The wire format follows the platform schema: memoryType/memoryScope as
explicit enum strings, visibility nested under shareConfig on requests
and flattened back via custom Deserialize on responses, and tag FQNs
wrapped in the TagLabel envelope (Manual/Confirmed/Classification).

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Adds the model layer for the Java memories namespace:

- MemoryType, MemoryScope, MemoryVisibility enums (with @jsonvalue
  serialization to camelCase wire values like "Preference",
  "EntityScoped", "Private")
- CreateContextMemoryRequest with builder + toApiMap() that re-nests
  visibility under shareConfig and wraps tags as TagLabels
  (labelType=Manual, state=Confirmed, source=Classification)
- ContextMemory POJO with a static fromMap() factory that flattens
  shareConfig.visibility back to a top-level visibility field
- MemorySearchHit + MemorySearchResults with MemorySearchResults
  fromOpenSearch() parser for the OpenSearch hits.total.value /
  hits.hits[i]._source/_score response shape

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Introduces an api/ package with one namespace class per entity domain:

- AgentsApi: list / create (resolves persona + ability names via the
  parent AISdk back-reference)
- BotsApi:  list / get
- PersonasApi: list / get / create
- AbilitiesApi: list / get
- MemoriesApi: list (paginated) / get / create / delete / search
  (hybrid NLQ search via /api/v1/hybrid/nlq/search), backed by two
  AISdkHttpClient instances — one rooted at /api/v1/contextCenter/
  memories for CRUD and one rooted at /api/v1 for search

Adds the supporting low-level helpers on AISdkHttpClient:

- A new constructor that accepts a base path so additional clients
  can target the memories and search endpoints
- Generic getMap / postMap / delete (path, params) methods used by
  the namespace classes for arbitrary JSON requests against any
  base URL

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Converts AISdk's per-entity flat methods to namespaced accessors that
return the new api/ package classes, matching the Python and
TypeScript SDK shape.

- AISdk now constructs three AISdkHttpClient instances (default,
  memories CRUD, hybrid search) and exposes agents() / bots() /
  personas() / abilities() / memories() accessors
- Removes flat methods listAgents / createAgent / listBots / getBot /
  listPersonas / getPersona / createPersona / listAbilities /
  getAbility — callers must use client.agents().list() etc.
- agent(name) factory and the no-arg agent() default-agent factory
  are unchanged
- AISdkTest and IntegrationTest are updated to the namespace form
- Adds MemoriesApiTest covering list pagination, primaryEntityFqn
  filtering, limit, get, create (toApiMap shape including tag
  wrapping and shareConfig nesting), delete (soft + hard), search
  parameters, search filter JSON serialization, OpenSearch response
  parsing, and ContextMemory.fromMap visibility flattening

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Adds the memory data model surface for the upcoming MemoriesApi namespace:
MemoryType / MemoryScope / MemoryVisibility string-literal unions,
ContextMemory, MemorySearchHit, MemorySearchResults, and
CreateContextMemoryRequest. Mirrors the Pydantic models in the Python
SDK and is purely additive — existing exports are unchanged.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Introduces a new src/api/ directory with one namespace class per
entity domain:

- AgentsApi: list / create (resolves persona + ability names via the
  parent AISdk back-reference)
- BotsApi: list / get
- PersonasApi: list / get / create
- AbilitiesApi: list / get
- MemoriesApi: list (paginated, optionally filtered by primaryEntityFqn)
  / get / create / delete (soft + hard) / search (hybrid NLQ via
  /api/v1/hybrid/nlq/search), backed by two HttpClient instances —
  one rooted at /api/v1/contextCenter/memories for CRUD and one at
  /api/v1 for the search endpoint.

Adds the supporting helpers on HttpClient: a generic delete() method,
boolean query-parameter values (for hardDelete), entityType /
entityName on public verb methods so namespaces can surface typed
404 errors, and a parseBody() helper that returns void for empty
responses.

The existing flat methods on AISdk remain in place for now; the next
commit replaces them with namespace accessors.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

…sors

Drops the listAgents / createAgent / listBots / getBot / listPersonas /
getPersona / createPersona / listAbilities / getAbility surface from
AISdk and replaces it with composed namespace fields:

  client.agents.list() / client.agents.create(req)
  client.bots.list()   / client.bots.get(name)
  client.personas.list() / client.personas.get(name) / client.personas.create(req)
  client.abilities.list() / client.abilities.get(name)
  client.memories.list() / .get / .create / .delete / .search

The agent handle factory (client.agent(name)) is unchanged.

AISdk now constructs one HttpClient per entity rooted at its API
path; the namespace classes issue plain relative requests against
that base URL. The legacy getAbsolute / postAbsolute helpers on
HttpClient are removed since no caller needs them anymore.

Tests: client.test.ts and integration.test.ts mechanically renamed
flat-method calls to namespace form. README.md updated to show the
namespace API and adds a Context Memories section.

This is a breaking change at v0.1.x (documented in CHANGELOG.md /
spec).

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Brainstorming/plan artifacts live alongside the code locally but are
not part of the merged history. Mirrors the existing docs/plans/**
ignore pattern.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Adds a new "Context Memories" section to:

- README.md (root) — concise pitch + Python/TS/Java/CLI examples for
  the create -> list -> search -> delete cycle, including filter usage.
- python/README.md — full examples (sync + async), enum cheatsheet,
  stored-fields table.
- java/README.md — namespace examples for all entities (replaces the
  old flat method calls that were missed in the Phase 3 refactor) plus
  a Memories section with builder usage and the stored-fields table.
- cli/README.md — `ai-sdk memories` subcommand reference with all
  flags, enum values, and a --json example.

The TypeScript README already had a Memories section from Phase 2.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Adds a synthetic "AskCollate (default)" entry as the first item in the
chat agent selection menu. Selecting it routes the chat to the
platform's default agent (use_default = true) instead of a named
dynamic agent. The menu is now useful even when there are no
API-enabled named agents.

Behaviour:
- Menu always has AskCollate at the top with description
  "The platform's default agent — no specific agent selected"
- Selecting AskCollate sets use_default=true and labels the header
  "AskCollate"
- Selecting a named agent sets use_default=false (clears the flag if
  the chat was previously in default mode)

Adds 4 unit tests covering the prepend, the empty-list case, and both
selection paths.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Two related TUI improvements:

1. Markdown links: previously the renderer fell into the catch-all
   _ => {} arm for Tag::Link / TagEnd::Link, so the visible label
   rendered with no styling and the URL was dropped. Now the label
   gets a blue + underlined style, and " (url)" is appended in
   dimmed text afterwards. Nested links are supported via a URL
   stack.

2. Thinking steps: the live-streaming and persisted thinking blocks
   were rendered as raw lines, so any markdown the agent emitted
   (links, bold, code) leaked through unstyled. They now go through
   a new render_markdown_dimmed() that runs the full markdown
   pipeline and then collapses every foreground colour to DarkGray
   while preserving bold/italic/underline modifiers. Each rendered
   line is indented two spaces under the "Thinking:" header.

Adds 3 unit tests: link label + url presence, link underline modifier,
and dimmed-renderer colour override + modifier preservation.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

The audit found two stale spots in docs and examples:

1. docs/API_REFERENCE.md — every per-SDK method table still showed
   the old flat methods (client.list_agents, client.listAgents,
   client.listAgents() in Java) and was missing the new memories
   namespace entirely. The Server Endpoint Map missed the memory
   CRUD and hybrid search endpoints. Data Models section had no
   ContextMemory / CreateContextMemoryRequest / MemorySearchResults /
   EntityReference. Feature Parity Matrix missed the Memories row
   and the default-agent feature.

2. cookbook/{gdpr-dsar-compliance,dbt-pr-review,dq-failure-slack-
   notifications}/agent-config.md — the TypeScript and Java code
   blocks still called client.createAgent(...). Updated to
   client.agents.create(...) for TS and client.agents().create(...)
   for Java.

Verified: a repo-wide grep for flat-method patterns now returns
zero hits outside the CHANGELOG migration table and gitignored
local design docs. All linters pass.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Long agent runs (>2 min) were being cut mid-stream because each SDK's
HTTP client applied its standard request timeout to SSE response bodies.
The failure surfaces as a decode/connection error, masking the timeout
as the real cause.

- Rust CLI: separate streaming reqwest client with no timeout
- Python: separate httpx client with read=None for post_stream (sync + async)
- TypeScript: dropped AbortController deadline from postStream
- Java: only connectTimeout was set; added Javadoc warnings to lock that in

Default timeout for non-streaming calls bumped 120 -> 900s across all four.

Also adds 'ai-sdk chat --debug [PATH]' to the Rust CLI: writes SSE event
traces (each event, message, and the underlying error + unparsed buffer
when a stream dies) to a file. Defaults to ~/.ai-sdk/chat-debug.log.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Minor bump: SSE timeout handling changed across all SDKs (separate
streaming client / no request deadline on streams), default request
timeout 120s -> 900s, and the Rust CLI gained `chat --debug [PATH]`
for SSE event tracing.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>