Skip to content

Changelog#

Unreleased#

  • Auto-retry Anthropic streams that end prematurely with empty responses, and auto-continue when content was partially received.

0.115.5#

  • Fix chat getting stuck when LLM streaming connection hangs or is cancelled.

0.115.4#

  • Fix auto-compact race condition where stale messages were sent to model after compaction, causing context overflow in subagents. #357

0.115.3#

  • Fix MCP server auth being invalidated when only URL query parameters change.
  • Fix MCP HTTP transport using stale auth token instead of reading latest from state, and retry initialization with token refresh on 403/4xx errors.
  • Fix MCP initialize request sending server config name as client name instead of "ECA".

0.115.2#

  • Fix STDIO MCP server deadlock caused by list_changed notification handlers blocking the reader thread.
  • Consider Anthropic internal server error as a retriable error.
  • Improve MCP error logging to show error code, message, and data instead of null.
  • Improve error handling when MCP promtps fail on server side.
  • Use rewrite-json to edit jsons without losing formatting.
  • Allow shell output redirections to /tmp/ in plan/explorer agents instead of denying them.
  • Fix crash when hook returns additionalContext and message has string content (e.g. /compact command).

0.115.1#

  • Improve OAuth callback error page to show error code, description, and error URI from the authorization server response.
  • Bump plumcp to 0.2.0-beta4, simplify MCP tool call error handling now that HTTP 400/404/500 are returned as JSON-RPC errors.
  • Fix MCP server incorrectly marked as running when HTTP transport returns 4xx (e.g. 401/403) during initialization.

0.115.0#

  • Improve native image size for smaller eca binaries
  • Drop UPX compression for all native image builds.
  • Optimize native image for size with -Os.
  • Add optional model and variant parameters to spawn_agent tool, allowing users to run subagents on a different model or variant than the current conversation.
  • Improve MCP OAuth spec compliance: add Protected Resource Metadata discovery (RFC 9728), OIDC Discovery 1.0 fallback for authorization server metadata, fix scope parameter formatting, include resource parameter (RFC 8707) in authorization and token requests, and support clientId in MCP server config for pre-registered OAuth applications.
  • Add trust mode: clients can send trust: true in chat/prompt to auto-accept tool calls that would require manual approval.
  • Improve MCP tool call error handling

0.114.2#

  • Fix completion requests failing with OpenAI subscription models by always streaming on the Responses API. eca-emacs#183
  • Disable upx for Windows binaries. #362

0.114.1#

  • Fix native image on Windows not running on older CPUs by enabling -march=compatibility. #362
  • Fix simple summaries in tool calls for ollama.
  • Add MCP server instructions to LLM context. #361

0.114.0#

  • Delete chats older than 7 days on server startup.
  • Use human-readable workspace cache directory names (e.g. my-project_a1b2c3d4), with automatic migration from old hash-only format.
  • Show MCP server URL in the details page and allow editing command/args/url inline with mcp/updateServer endpoint.
  • Add mcp/updateServer to support change mcp commands/urls from client UI.

0.113.1#

  • Fix MCP server threads blocking ECA shutdown when stuck during initialization; startup now uses daemon threads with interrupt-based cancellation for clean exit.
  • Re-initialize MCP session and retry tool call on HTTP 404 (session expired) or 5xx (e.g. pod swap dropping the SSE stream), per MCP Streamable HTTP spec.
  • Fix preRequest hook all_messages containing only the current turn instead of the full conversation history.

0.113.0#

  • Clarify Task tool prompt for task ordering, task granularity, concurrent starts, and clearing finished task lists.
  • Improve MCP server auth: stop auto-opening browser on OAuth, add mcp/connectServer request for client-driven auth and mcp/logoutServer notification for clearing credentials.
  • Replace MCP Java SDK with plumcp for MCP client communication. SSE transport is no longer supported (deprecated in MCP spec 2025-03-26).
  • Fix crash on invalid YAML frontmatter in plugin skill files.
  • Show "Waiting subagent" instead of "Calling tool" in progress messages during subagent execution.

0.112.1#

  • Fix MCP OAuth credentials cache not invalidating when the server URL changes.
  • Add isSubagent condition variable for chat system instructions
  • Breaking: Replace bodyPattern with errorPattern in retryRules, which matches against any error text (response body, error message, or exception message).
  • Retry on "Remote host terminated the handshake" TLS errors.
  • Fix empty "Error: " message on connection failures (e.g. DNS resolution, connection refused) and retry them as transient errors.

0.112.0#

  • Add plugins config for loading external configuration from git repos or local paths. #349
  • Plugins can provide skills, MCP servers, agents, commands, hooks, rules, and arbitrary config overrides.
  • Add commands /plugins, /plugin-install, and /plugin-uninstall.
  • Add official plugin marketplace as built-in plugin source (eca), available by default at plugins.eca.dev.
  • Fix race condition where stopping a prompt and immediately sending a new one could cause two concurrent prompts with no way to stop the older one.
  • Support inherit key in agent config to inherit settings from another agent.

0.111.0#

  • Fix MCP server initialization crash (String cannot be cast to IPersistentCollection) when OAuth metadata endpoint returns a non-JSON or error response.
  • Auto-approve eca__read_file for tool call output cache files (~/.cache/eca/toolCallOutputs).
  • Support configurable retry rules per provider via retryRules, allowing users to define custom HTTP status and/or body pattern matching for automatic retries with optional labels.

0.110.3#

  • Fix outputTruncation.sizeKb not being honored
  • Improve tasks feature prompt

0.110.2#

  • Fix MCP OAuth client registration failing for servers that require Content-Type header
  • Prevent MCP servers from getting stuck at "starting" when OAuth or other initialization errors occur.
  • Fix GitHub Copilot premium request overconsumption:
  • Chat titles LLM request don't spend premium requests.
  • Subagents now use same premium request of primary agent.
  • gpt-5.3-codex and gpt-5.4 now spend way less premium requests.

0.110.1#

  • Fix MCP Streamable HTTP servers stuck at starting due to Java SDK DummyEvent bug on 202 notification responses, and spurious OAuth detection from proxies returning 401 without www-authenticate on HEAD requests.

  • Retry MCP server initialization on transient HTTP errors (e.g. chunked encoding EOF from infrastructure closing SSE connections).

  • Add support for gpt-5.4 variants

0.110.0#

  • Fix rollback when there are subagents messages in chat.
  • Add Task tool #246
  • Use OpenAI Responses API for GitHub Copilot models that require it (gpt-5.3-codex, gpt-5.4).

0.109.6#

  • Allow manually starting MCP servers even when configured with disabled: true (still not auto-started).
  • Fix MCP OAuth discovery for servers that don't support HEAD requests (e.g. Glean) by falling back to POST, and fix two-step metadata discovery (PRM → Authorization Server Metadata per RFC 8414).

0.109.5#

  • Fix clear messages to reset usage tokens as well.

0.109.4#

  • Fix clear messages to reset last api used as well.
  • Avoid server stuck during initialize due to threads unavailable. #169

0.109.3#

  • Fix invalid request 400, for tool calls with nil arguments failing when using Anthropic models via GitHub Copilot. #340

0.109.2#

  • Add chat/clear message.

0.109.1#

  • Strip markdown code fences from rewrite results via new rewrite/contentReceived replace content type. #336

0.109.0#

  • Improve /login text.
  • Add support to workspace folders change via git worktree via workspace/didChangeWorkspaceFolders.

0.108.1#

  • Support other than text type of tool responses in MCPs. #331

0.108.0#

  • Add support for auto retry for http codes: 429, 500, 502, 503 and 529. #224

0.107.0#

  • Fix home (~) expand for command paths in config
  • Add Context overflow recovery feature.

0.106.0#

  • Change Openai summary from detailed -> auto.
  • Add variants feature #302
  • By default no variant is set, using model default.
  • Some models (openai/anthropic) will have built-in variants (low, medium, high, max etc).
  • Users can create its own variants for their providers/models.

0.105.0#

  • Add network configuration for custom CA certificates and mTLS client certificates. #327
  • New network config key with caCertFile, clientCert, clientKey, and clientKeyPassphrase.
  • Environment variable fallbacks: SSL_CERT_FILE, NODE_EXTRA_CA_CERTS, ECA_CLIENT_CERT, ECA_CLIENT_KEY, ECA_CLIENT_KEY_PASSPHRASE.
  • Custom CA certificates are additive to the JVM default trust store.

0.104.2#

  • Avoid showing empty chats in resume. #326
  • Fix anthropic adaptive thinking option causing invalid request errors.
  • Fix anthropic web server search corner case causing invalid request errors.

0.104.1#

  • Improve fetch models logging when failling. #313

0.104.0#

  • Allow disable chat title generation via chat title as false. #322

0.103.2#

  • Support dynamic string content in markdown configs (agents, skills). #317

0.103.1#

  • Improve /doctor command to show clearly model used + login providers.
  • Fix disabled mcp tools not showing properly in mcp details UI.

0.102.0#

  • Fix directory_tree returning empty results for paths outside workspace folders.
  • Fix anthropic "invalid_request_error" that may happen when doing server web-search + thinking, causing errors in chat.
  • Allow fetching models from providers with no api keys required. #313
  • Fix usage tokens for anthropic when doing server web-searches. #307
  • Show server websearches as tool calls.
  • Improve plan/explorer delegation guidance and reduce friction for common read-only investigation workflows.
  • Make anthropic/claude-sonnet-4-6 default model for anthropic.

0.101.1#

  • Allow non JWT tokens for custom providers with openai api. #303
  • Fix NPE when first time setting up ECA with no model. #304
  • Minor explorer/tool call truncation prompt improvements.

0.101.0#

  • Foreground subagents support! #160

0.100.4#

  • Avoid tool call errors on anthropic api.
  • fix model capabilities not being set properly for some custom models.

0.100.3#

  • Improve regex of denied commands in plan agent.
  • Fix fetch models regression. #299

0.100.2#

  • Bump MCP java sdk to 0.17.2.
  • Avoid duplicated sonnet-4-5 model.

0.100.1#

  • Improve anthropic login wording.
  • Fix regression in check for available models. #297

0.100.0#

  • Add config.json schema. #293
  • Use models.dev as the only dynamic model source for configured providers, with fetchModels: false as an opt-out to use only static config models. #292
  • Deprecate behavior in favor of agent.

0.99.0#

  • Truncate tool call outputs automatically to avoid hit context size limit. #284
  • Save output to eca cache folde and tell LLM where the full content is.
  • Add toolCall outputTruncation config to customize size or lines.
  • Improve shell command summary by stripping cd <workspace-root> && prefix.

0.98.5#

  • Add openai via subscription gpt-5.3-codex model.

0.98.4#

  • Add claude-opus-4-6 model.

0.98.3#

  • Fix "approve and remember" tool call when checking files outside workspace.
  • Improve MCP server shutdown. #287
  • Improve grep tool to support different output modes. #289

0.98.2#

  • Lower default autoCompactPercentage from 85 -> 75.
  • Add x-initiator to github-copilot requests, improving the usage of premium requests counter. #138

0.98.1#

  • Fix openai-chat tool calls freezing when providers emit duplicate/invalid tool_calls[].id values.

0.98.0#

  • Add support for adding extraHeaders to models configuration.

0.97.7#

  • Remove the need to use codex instructions for codex sub, replacing with ECA prompt like all other providers.

0.97.6#

  • Improve Anthropic API to support more models like kimi-k2.5.

0.97.5#

  • Support moonshot models via anthropic API.

0.97.4#

  • Fix openai request when using non codex models but using openai subscription.

0.97.3#

  • Fix non sse remote mcps not working.

0.97.2#

  • Do not try to oauth remote mcp servers if they have Authorization header.

0.97.1#

  • Fix regression in last release MCPs not being started properly.

0.97.0#

  • Support update list of mcp tools when requested from server via list_changed notification.

0.96.1#

  • Fix /resume duplicating chats. #278
  • Reduce noise in auto-compaction, not mentioning the summary. #280

0.96.0#

  • Add /skill-create command to create a ECA skill from a prompt. #277

0.95.0#

  • (OpenAI Chat) - Configurable reasoning history via reasoningHistory (model-level, default: all)
  • Fix exception in autocompact that could happen some times.

0.94.2#

  • Fix autocompact not cleaning tokens in memory and thinking it should auto compact again.

0.94.1#

  • Fix tools prompt override not working via config.

0.94.0#

  • Mention in chat some missing Anthropic responses errors when happens.
  • Auto compact via percentage defined in autoCompactPercentage. #257
  • Do not remove conversation history visually when compacting chat.

0.93.2#

  • Fix /compact removing chat history when prompt is stopped or some error happens. #142

0.93.1#

  • Fix chat title generation regression.

0.93.0#

  • Improve copilot login to mention to enable model at Copilot settings page.
  • New config API for prompts:
  • Support override any tool description via prompts tools <toolName> #271
  • Support override /init system prompt via prompts init
  • Support override /compact system prompt via prompts compact
  • Support override chat title system prompt via prompts chatTitle
  • Deprecate systemPrompt in favor of prompts chat
  • Deprecate completion systemPrompt in favor of prompts completion
  • Deprecate rewrite systemPrompt in favor of prompts rewrite

0.92.3#

  • Improve error handling on chat messages, avoiding stuck and losing chat history. #272
  • Fix directories not working in user prompt. #273

0.92.2#

  • Fix whitespace handle in uris. #270

0.92.1#

  • Add x-llm-application-name: eca to prompt requests, useful to track and get metrics when using LLM gateways.

0.92.0#

  • Fix Gemini (OpenAI compatible). #247
  • Improve login wording + add cancel option. #205

0.91.2#

  • Fix eca_shell_command to include stderr output even when exit 0.

0.91.1#

  • Fix openai pro login exceptions related to graalvm.

0.91.0#

  • Codex login support via /login openai and selecting pro option. #261
  • Add gpt-5.2-codex model.

0.90.1#

  • Fix tokens not renewing between tool calls. #258
  • Fix diff correct numbers in added/removed. #259
  • Fix post tool call hook trigger
  • Fix Gemini tool calling. #247

0.90.0#

  • Skills support following https://agentskills.io . #241
  • Make skills available via commands as well e.g. /my-skill.
  • new command /skills to list available skills.

0.89.0#

  • Support http MCP servers that require oauth. #51
  • Add basic username/password proxy authentication support and recognize lowercase http[s]_proxy env var alongside HTTP[S]_PROXY. #248
  • Avoid tool call of invalid names

0.88.0#

  • Add dynamic model discovery via fetchModels provider config for OpenAI-compatible /models endpoints
  • Improve error handling for incompatible models messages in chat. #209
  • Support server__tool_name in disabledTools config as well.
  • Fix clojure-mcp regression where ECA could not edit files via clojure-mcp even reading before using its tools.

0.87.2#

  • Fix openai-chat tool call + support for Mistral API #233
  • Skip missing/unreadable @file references when building context
  • Fix regression: /compact not working for some models. Related to #240

0.87.1#

  • Improve read-file summary to show final range properly.
  • Improve model capabilities for providers which model name has slash: my-provider/anthropic/my-model

0.87.0#

  • Support Google Gemini thought signatures.
  • Support gemini-3-pro-preview and gemini-3-flash-preview models in Google and Copilot providers.
  • Fix deepseek reasoning with openai-chat API #228
  • Support ~ in dynamic string parser.
  • Support removing nullable values from LLM request body if the value in extraPayload is null. #232

0.86.0#

  • Improve agent behavior prompt to mention usage of editor_diagnostics tool. #230
  • Use selmer syntax for prompt templates.

0.85.3#

  • Support openai/gpt-5.2 and github-copilot/gpt-5.2 by default.

0.85.2#

  • Support providers <provider> httpClient version config, allowing to use http-1.1 for some providers like lmstudio. #229

0.85.1#

  • Fix backwards compatibility for chat rollback.

0.85.0#

  • Enhanced hooks documentation with new types (sessionStart, sessionEnd, chatStart, chatEnd), JSON input/output schemas, execution options (timeout)

  • Fix custom tools to support argument numbers.

  • Improve read_file summary to mention offset being read.
  • Enhanced hooks documentation with new types (sessionStart, sessionEnd, chatStart, chatEnd), JSON input/output schemas, execution options (timeout)
  • Support rollback only messages, tool call changes or both in chat/rollback.

0.84.2#

  • Fix ${netrc:...} to consider :netrcFile config properly.

0.84.1#

  • Fix ${netrc:...} to consider :netrcFile config.

0.84.0#

  • Improve /compact UI in chat after running, cleaning chat and showing the new summary.
  • Better config values dynamic string parse:
    • Support ${classapath:path/to/eca/classpath/file} in dynamic string parse.
    • Support ${netrc:api.foo.com} in dynamic string parse to parse keys. #200
    • Support default env values in ${env:MY_ENV:default-value}.
    • Support for ECA_CONFIG and custom config file.
  • Deprecate configs:
  • systemPromptFile in favor of systemPrompt using ${file:...} or ${classpath:...}
  • urlEnv in favor of url using ${env:...}
  • keyEnv in favor of key using ${env:...}
  • keyRc in favor of key using ${netrc:...}
  • compactPromptFile in favor of compactPrompt using ${classpath:...}

0.83.0#

  • Support dynamic string parse (${file:/path/to/something} and ${env:MY_ENV}) in all configs with string values. #200

0.82.1#

  • Fix custom tools output to return stderr when tool error. #219

0.82.0#

  • Support nested folder for rules and commands. #220

0.81.0#

  • Support rollback file changes done by write_file, edit_file and move_file. #218
  • Improve rollback to keep consistent UI before the rollback, fixing tool names and user messages.

0.80.4#

  • Fix binary for macos amd64. #217

0.80.3#

  • Fix release

0.80.2#

  • Update anthropic default models to include opus-4.5
  • Update anthropic default models to use alias names.

0.80.1#

  • Add new models to GitHub config (Gpt 5.1 and Opus 4.5).

0.80.0#

  • Add support to rollback messages via chat/rollback and chat/clear messages. #42

0.79.1#

  • Improve system prompt to add project env context.

0.79.0#

  • Fix absolute paths being interpreted as commands. #199
  • Remove non used sync models code during initialize. #100
  • Fix system prompt to mention the user workspace roots.

0.78.4#

  • Fix regression exceptions on specific corner cases with log obfuscation.

0.78.3#

  • Add openai/gpt-5.1 to default models.

0.78.2#

  • Add workspaces to /doctor
  • Improve LLM request logs to include headers.

0.78.1#

  • Fix regression: completion broken after rewrite feature API changes.

0.78.0#

  • Prefix tool name with server to LLM: __. #196
  • Remove eca_ prefix from eca tools, we already pass server prefix (eca) after #196.
  • Add approval arg to preToolCall hook input.
  • Add error rewrite message content message.

0.77.1#

  • Fix token renew when using rewrite feature.
  • Improve rewrite error handling.

0.77.0#

  • Custom providers do not require the existense of key or keyEnv. #194
  • New feature: rewrite. #13

0.76.0#

  • Updated instructions for /login command and invalid input handling.
  • Fix server name on chat/contentReceived when preparing tool call.
  • Fix variable replacing in some tool prompts.
  • Improve planning mode prompt and tool docs; clarify absolute-path usage and preview rules.
  • Centralize path approval for tools and always list all missing required params in INVALID_ARGS.

0.75.4#

  • Fix 0.75.3 regression on custom openai-chat providers.

0.75.3#

  • Support custom think tag start and end for openai-chat models via think-tag-start and think-tag-end provider configs. #188
  • Bump MCP java sdk to 0.15.0.

0.75.2#

  • Add missing models supported by Github Copilot
  • Fix regression: openai-chat tool call arguments error on some models.

0.75.1#

  • Improve protocol for tool call output formatting for tools that output json.
  • Fix inconsistencies in eca_read_file not passing correct content to LLM when json.

0.75.0#

  • Improved file contexts: now use :lines-range
  • BREAKING ECA now only supports standard plain-text netrc as credential file reading. Drop authinfo and gpg decryption support. Users can choose to pass in their own provisioned netrc file from various secure source with :netrcFile in ECA config.

0.74.0#

  • Improved eca_edit_file to automatically handle whitespace and indentation differences in single-occurrence edits.
  • Fix contexts in user prompts (not system contexts) not parsing lines ranges properly.
  • Support non-stream providers on openai-chat API. #174

0.73.5#

  • Support use API keys even if subscription is logged. #175

0.73.4#

  • Fix tool call approval ignoring eca tools.

0.73.3#

  • Fix tool call approval ignoring configs for mcp servers.

0.73.2#

  • Fix tool call approval thread lock.

0.73.1#

  • Improve chat title generation.
  • Fix completion error handling.
  • Default to openai/gpt-4.1 on completion.

0.73.0#

  • Add :config-file cli option to pass in config.
  • Add support for completion. #12

0.72.2#

  • Run preToolCall hook before user approval if any. #170

0.72.1#

  • Only include parallel_tool_calls to openai-responses and openai-chat if true. #169

0.72.0#

  • Support clojureMCP dry-run flags for edit/write tools, being able to show preview of diffs before running tool.

0.71.3#

  • Assoc parallel_tool_calls to openai-chat only if truth.

0.71.2#

  • Fix regression in /compact command. #162
  • Fix to use local zone for time presentation in /resume.

0.71.1#

  • Use web-search false if model capabiltiies are not found.

0.71.0#

  • Support /resume a specific chat.

0.70.6#

  • Fix openai-chat api not following completionUrlRelativePath.

0.70.5#

  • Fix web-search not working for custom models using openai/anthropic apis.

0.70.4#

  • Support visible field in hooks configuration to show or not in client.

0.70.3#

  • Deprecate prePrompt and postPrompt in favor of preRequest and prePrompt.

0.70.2#

  • Fix model capabilities for models with custom names.

0.70.1#

  • Fix prePrompt hook.

0.70.0#

  • Add hooks support. #43

0.69.1#

  • Fix regression on models with no extraPayload.

0.69.0#

  • Support multiple model configs with different payloads using same model name via modelName config. (Ex: gpt-5 and gpt-5-high but both use gpt-5)

0.68.1#

  • Add anthropic/haiku-4.5 model by default.

0.68.0#

  • Unwrap mentioned @contexts in prompt appending as user message its content. #154

0.67.0#

  • Improved flaky test #150
  • Obfuscate env vars in /doctor.
  • Bump clj-otel to 0.2.10
  • Rename $ARGS to $ARGUMENTS placeholder alias for custom commands.
  • Support recursive AGENTS.md file inclusions with @file mention. #140

0.66.1#

  • Improve plan behavior prompt. #139

0.66.0#

  • Add support for secrets stored in authinfo and netrc files
  • Added tests for stopping concurrent tool calls. #147
  • Improve logging.
  • Improve performance of chat/queryContext.

0.65.0#

  • Added ability to cancel tool calls. Only the shell tool currently. #145
  • Bump mcp java sdk to 0.14.1.
  • Improve json output for tools that output json.

0.64.1#

  • Fix duplicated arguments on toolCallPrepare for openai-chat API models. https://github.com/editor-code-assistant/eca-emacs/issues/56

0.64.0#

  • Add server to tool call messages.

0.63.3#

  • Fix last word going after tool call for openai-chat API.

0.63.2#

  • Fix retrocompatibility with some models not working with openai-chat like deepseek.

0.63.1#

  • Add gpt-5-codex model as default for openai provider.

0.63.0#

  • Support "accept and remember" tool call per session and name.
  • Avoid generating huge chat titles.

0.62.1#

  • Add claude-sonnet-4.5 for github-copilot provider.
  • Add prompt-received metric.

0.62.0#

  • Use a default of 32k tokens for max_tokens in openai-chat API.
  • Improve rejection prompt for tool calls.
  • Use max_completion_tokens instead of max_tokens in openai-chat API.
  • Support context/tokens usage/cost for openai-chat API.
  • Support anthropic/claude-sonnet-4.5 by default.

0.61.1#

  • More tolerant whitespace handling after data:.
  • Fix login for google provider. #134

0.61.0#

  • Fix chat titles not working for some providers.
  • Enable reasoning for google models.
  • Support reasoning blocks in models who use openai-chat api.

0.60.0#

  • Support google gemini as built-in models. #50

0.59.0#

  • Deprecate repoMap context, will be removed in the future.
  • After lots of tunnings and improvements, the repoMap is no longer relevant as eca_directory_tree provides similar and more specific view for LLM to use.
  • Support toolCall shellCommand summaryMaxLength to configure UX of command length. #130

0.58.2#

  • Fix MCP prompt for native image.

0.58.1#

  • Improve progress notification when tool is running.

0.58.0#

  • Bump MCP java sdk to 0.13.1
  • Improve MCP logs on stderr.
  • Support tool call rejection with reasons inputed by user. #127

0.57.0#

  • Greatly reduce token consuming of eca_directory_tree
  • Ignoring files in gitignore
  • Improving tool output for LLM removing token consuming chars.

0.56.4#

  • Fix renew oauth tokens when it expires in the same session.

0.56.3#

  • Fix metrics exception when saving to db.

0.56.2#

  • Fix db exception.

0.56.1#

  • Fix usage reporting.

0.56.0#

  • Return new chat metadata content.
  • Add chat title via prompt to LLM.

0.55.0#

  • Add support for OpenTelemetry via otlp config.
  • Export metrics of server tasks, tool calls, prompts, resources.

0.54.4#

  • Use jsonrpc4clj instead of lsp4clj.
  • Bump graalvm to 24 and java to 24 improving native binary perf.

0.54.3#

  • Avoid errors on multiple same MCP server calls in parallel.

0.54.2#

  • Fix openai cache tokens cost calculation.

0.54.1#

  • Improve welcome message.

0.54.0#

  • Improve large file handling in read-file tool:
  • Replace basic truncation notice with detailed line range information and next-step instructions.
  • Allow users to customize default line limit through tools.readFile.maxLines configuration (keep the current 2000 as default).
  • Moved the future in :on-tools-called and stored it in the db. #119
  • Support compactPromptFile config.
  • Fix tools not being listed for servers using mcp-remote.

0.53.0#

  • Add /compact command to summarize the current conversation helping reduce context size.
  • Add support for images as contexts.

0.52.0#

  • Support http-streamable http servers (non auth support for now)
  • Fix promtps that send assistant messages not working for anthropic.

0.51.3#

  • Fix manual anthropic login to save credentials in global config instead of cache.

0.51.2#

  • Minor log improvement of failed to start MCPs.

0.51.1#

  • Bump mcp java sdk to 1.12.1.
  • Fix mcp servers default timeout from 20s -> 60s.

0.51.0#

  • Support timeout on eca_shell_command with default to 1min.
  • Support @cursor context representing the current editor cursor position. #103

0.50.2#

  • Fix setting the web-search capability in the relevant models
  • Fix summary text for tool calls using openai-chat api.

0.50.1#

  • Bump mcp-java-sdk to 0.12.0.

0.50.0#

  • Added missing parameters to toolCallRejected where possible. PR #109
  • Improve plan prompt present plan step.
  • Add custom behavior configuration support. #79
  • Behaviors can now define defaultModel, disabledTools, systemPromptFile, and toolCall approval rules.
  • Built-in agent and plan behaviors are pre-configured.
  • Replace systemPromptTemplateFile with systemPromptFile for complete prompt files instead of templates.
  • Remove nativeTools configuration in favor of toolCall approval and disabledTools.
  • Native tools are now always enabled by default, controlled via disabledTools and toolCall approval.

0.49.0#

  • Add totalTimeMs to reason and toolCall content blocks.

0.48.0#

  • Add nix flake build.
  • Stop prompt does not change the status of the last running toolCall. #65
  • Add toolCallRunning content to chat/contentReceived.

0.47.0#

  • Support more providers login via /login.
  • openai
  • openrouter
  • deepseek
  • azure
  • z-ai

0.46.0#

  • Remove the need to pass requestId on prompt messages.
  • Support empty /login command to ask what provider to login.

0.45.0#

  • Support user configured custom tools via customTools config. #92
  • Fix default approval for read only tools to be allow instead of ask.

0.44.1#

  • Fix renew token regression.
  • Improve error feedback when failed to renew token.

0.44.0#

  • Support deny tool calls via toolCall approval deny setting.

0.43.1#

  • Safely rename default* -> select* in config/updated.

0.43.0#

  • Support chat/selectedBehaviorChanged client notification.
  • Update models according with supported models given its auth or key/url configuration.
  • Return models only authenticated or logged in avoid too much models on UI that won't work.

0.42.0#

  • New server notification config/updated used to notify clients when a relevant config changed (behaviors, models etc).
  • Deprecate info inside initialize response, clients should use config/updated now.

0.41.0#

  • Improve anthropic extraPayload requirement when adding models.
  • Add message to when config failed to be parsed.
  • Fix context completion for workspaces that are not git. #98
  • Fix session tokens calculation.

0.40.0#

  • Drop agentFileRelativePath in favor of behaviors customizations in the future.
  • Unwrap chat config to be at root level.
  • Fix token expiration for copilot and anthropic.
  • Considerably improve toolCall approval / permissions config.
  • Now with thave multiple optiosn to ask or allow tool calls, check config section.

0.39.0#

  • Fix session-tokens in usage notifications.
  • Support context limit on usage notifications.
  • Fix session/message tokens calculation.

0.38.3#

  • Fix anthropic token renew.

0.38.2#

  • Fix command prompts to allow args with spaces between quotes.
  • Fix anthropic token renew when expires.

0.38.1#

  • Fix graalvm properties.

0.38.0#

  • Improve plan-mode (prompt + eca_preview_file_change tool) #94
  • Add fallback for matching / editing text in files #94

0.37.0#

  • Require approval for eca_shell_command if running outside workspace folders.
  • Fix anthropic subscription.

0.36.5#

  • Fix pricing for models being case insensitive on its name when checking capabilities.

0.36.4#

  • Improve api url error message when not configured.

0.36.3#

  • Fix anthropic/claude-3-5-haiku-20241022 model.
  • Log json error parsing in configs.

0.36.2#

  • Add login providers and server command to /doctor.

0.36.1#

  • Improved the eca_directory_tree tool. #82

0.36.0#

  • Support relative contexts additions via ~, ./ ../ and /. #61

0.35.0#

  • Anthropic subscription support, via /login anthropic command. #57

0.34.2#

  • Fix copilot requiring login in different workspaces.

0.34.1#

  • Fix proxy exception. #73

0.34.0#

  • Support custom UX details/summary for MCP tools. #67
  • Support clojureMCP tools diff for file changes.

0.33.0#

  • Fix reasoning titles in thoughts blocks for openai-responses.
  • Fix hanging LSP diagnostics requests
  • Add lspTimeoutSeconds to config
  • Support HTTP_PROXY and HTTPS_PROXY env vars for LLM request via proxies. #73

0.32.4#

  • Disable eca_plan_edit_file in plan behavior until better idea on what plan behavior should do.

0.32.3#

  • Consider AGENTS.md instead of AGENT.md, following the https://agents.md standard.

0.32.2#

  • Fix option to set default chat behavior from config via chat defaultBehavior. #71

0.32.1#

  • Fix support for models with / in the name like Openrouter ones.

0.32.0#

  • Refactor config for better UX and understanding:
  • Move models to inside providers.
  • Make customProviders compatible with providers. models need to be a map now, not a list.

0.31.0#

  • Update copilot models
  • Drop uneeded ollama useTools and ollama think configs.
  • Refactor configs for config providers unification.
  • <provider>ApiKey and <providerApiUrl> now live in :providers "<provider>" :key.
  • Move defaultModel config from customProvider to root.

0.30.0#

  • Add /login command to login to providers
  • Add Github Copilot models support with login.

0.29.2#

  • Add /doctor command to help with troubleshooting

0.29.1#

  • Fix args streaming in toolCallPrepare to not repeat the args. https://github.com/editor-code-assistant/eca-nvim/issues/28

0.29.0#

  • Add editor tools to retrieve information like diagnostics. #56

0.28.0#

  • Change api for custom providers to support openai-responses instead of just openai, still supporting openai only.
  • Add limit to repoMap with default of 800 total entries and 50 per dir. #35
  • Add support for OpenAI Chat Completions API for broad third-party model support.
  • A new openai-chat custom provider api type was added to support any provider using the standard OpenAI /v1/chat/completions endpoint.
  • This enables easy integration with services like OpenRouter, Groq, DeepSeek, Together AI, and local LiteLLM instances.

0.27.0#

  • Add support for auto read AGENT.md from workspace root and global eca dir, considering as context for chat prompts.
  • Add /prompt-show command to show ECA prompt sent to LLM.
  • Add /init command to ask LLM to create/update AGENT.md file.

0.26.3#

  • breaking: Replace configs ollama host and ollama port with ollamaApiUrl.

0.26.2#

  • Fix chat/queryContext to not return already added contexts
  • Fix some MCP prompts that didn't work.

0.26.1#

  • Fix anthropic api for custom providers.
  • Support customize completion api url via custom providers.

0.26.0#

  • Support manual approval for specific tools. #44

0.25.0#

  • Improve plan-mode to do file changes with diffs.

0.24.3#

  • Fix initializationOptions config merge.
  • Fix default claude model.

0.24.2#

  • Fix some commands not working.

0.24.1#

  • Fix build

0.24.0#

  • Get models and configs from models.dev instead of hardcoding in eca.
  • Allow custom models addition via models <modelName> config.
  • Add /resume command to resume previous chats.
  • Support loading system prompts from a file.
  • Fix model name parsing.

0.23.1#

  • Fix openai reasoning not being included in messages.

0.23.0#

  • Support parallel tool call.

0.22.0#

  • Improve eca_shell_command to handle better error outputs.
  • Add summary for eca commands via summary field on tool calls.

0.21.1#

  • Default to gpt-5 instead of o4-mini when openai-api-key found.
  • Considerably improve eca_shell_command to fix args parsing + git/PRs interactions.

0.21.0#

  • Fix openai skip streaming response corner cases.
  • Allow override payload of any LLM provider.

0.20.0#

  • Support custom commands via md files in ~/.config/eca/commands/ or .eca/commands/.

0.19.0#

  • Support claude-opus-4-1 model.
  • Support gpt-5, gpt-5-mini, gpt-5-nano models.

0.18.0#

  • Replace chat behavior with plan.

0.17.2#

  • fix query context refactor

0.17.1#

  • Avoid crash MCP start if doesn't support some capabilities.
  • Improve tool calling to avoid stop LLM loop if any exception happens.

0.17.0#

  • Add /repo-map-show command. #37

0.16.0#

  • Support custom system prompts via config systemPromptTemplate.
  • Add support for file change diffs on eca_edit_file tool call.
  • Fix response output to LLM when tool call is rejected.

0.15.3#

  • Rename eca_list_directory to eca_directory_tree tool for better overview of project files/dirs.

0.15.2#

  • Improve eca_edit_file tool for better usage from LLM.

0.15.1#

  • Fix mcp tool calls.
  • Improve eca filesystem calls for better tool usage from LLM.
  • Fix default model selection to check anthropic api key before.

0.15.0#

  • Support MCP resources as a new context.

0.14.4#

  • Fix usage miscalculation.

0.14.3#

  • Fix reason-id on openai models afecting chat thoughts messages.
  • Support openai o models reason text when available.

0.14.2#

  • Fix MCPs not starting because of graal reflection issue.

0.14.1#

  • Fix native image build.

0.14.0#

  • Support enable/disable tool servers.
  • Bump mcp java sdk to 0.11.0.

0.13.1#

  • Improve ollama model listing getting capabilities, avoiding change ollama config for different models.

0.13.0#

  • Support reasoning for ollama models that support think.

0.12.7#

  • Fix ollama tool calls.

0.12.6#

  • fix web-search support for custom providers.
  • fix output of eca_shell_command.

0.12.5#

  • Improve tool call result marking as error when not expected output.
  • Fix cases when tool calls output nothing.

0.12.4#

  • Add chat command type.

0.12.3#

  • Fix MCP prompts for anthropic models.

0.12.2#

  • Fix tool calls

0.12.1#

  • Improve welcome message.

0.12.0#

  • Fix openai api key read from config.
  • Support commands via /.
  • Support MCP prompts via commands.

0.11.2#

  • Fix error field on tool call outputs.

0.11.1#

  • Fix reasoning for openai o models.

0.11.0#

  • Add support for file contexts with line ranges.

0.10.3#

  • Fix openai max_output_tokens message.

0.10.2#

  • Fix usage metrics for anthropic models.

0.10.1#

  • Improve eca_read_file tool to have better and more assertive descriptions/parameters.

0.10.0#

  • Increase anthropic models maxTokens to 8196
  • Support thinking/reasoning on models that support it.

0.9.0#

  • Include eca as a server with tools.
  • Support disable tools via config.
  • Improve ECA prompt to be more precise and output with better quality

0.8.1#

  • Make generic tool server updates for eca native tools.

0.8.0#

  • Support tool call approval and configuration to manual approval.
  • Initial support for repo-map context.

0.7.0#

  • Add client request to delete a chat.

0.6.1#

  • Support defaultModel in custom providers.

0.6.0#

  • Add usage tokens + cost to chat messages.

0.5.1#

  • Fix openai key

0.5.0#

  • Support custom LLM providers via config.

0.4.3#

  • Improve context query performance.

0.4.2#

  • Fix output of errored tool calls.

0.4.1#

  • Fix arguments test when preparing tool call.

0.4.0#

  • Add support for global rules.
  • Fix origin field of tool calls.
  • Allow chat communication with no workspace opened.

0.3.1#

  • Improve default model logic to check for configs and env vars of known models.
  • Fix past messages sent to LLMs.

0.3.0#

  • Support stop chat prompts via chat/promptStop notification.
  • Fix anthropic messages history.

0.2.0#

  • Add native tools: filesystem
  • Add MCP/tool support for ollama models.
  • Improve ollama integration only requiring ollama serve to be running.
  • Improve chat history and context passed to all LLM providers.
  • Add support for prompt caching for Anthropic models.

0.1.0#

  • Allow comments on json configs.
  • Improve MCP tool call feedback.
  • Add support for env vars in mcp configs.
  • Add mcp/serverUpdated server notification.

0.0.4#

  • Add env support for MCPs
  • Add web_search capability
  • Add o3 model support.
  • Support custom API urls for OpenAI and Anthropic
  • Add --log-level <level> option for better debugging.
  • Add support for global config file.
  • Improve MCP response handling.
  • Improve LLM streaming response handler.

0.0.3#

  • Fix ollama servers discovery
  • Fix .eca/config.json read from workspace root
  • Add support for MCP servers

0.0.2#

  • First alpha release

0.0.1#