Skip to content

Changelog#

Unreleased#

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#